Audio processing

[0001] The present invention relates to audio processing and methods and apparatus for audio processing.

[0002] Equalization of audio data is well known. Audio equalizers generally fall into one of three categories: (a) tone controls; (b) parametric equalizers; and (c) graphic equalizers. These categories are well-known and are therefore discussed only briefly below.

[0003] Tone controls use very small amounts of processing to raise or lower the audio frequency response that results from processing by a device. The increase or decrease of the frequency response is usually limited to two, or sometimes three, very broad bands of frequencies that are typically referred to as “Bass”, “Mid” and “Treble”. Tone controls are generally provided to allow an individual listener to control, with basic adjustments, the frequency response of the device to match their personal preference.

[0004] Parametric equalizers use a small amount of processing to generate effects such as simple bumps, dips, shelves, and high and low cut filters to modify sounds, often during audio recording and mixing. These individual audio effects (or modifications to the frequency response of the audio processing) are generally controllable by a number of continuously variable “parameters” (hence the name, “parametric equalizer”). For example, a bump effect is likely to be controllable by the following parameters: (i) gain, i.e. the amount of increase or decrease in frequency response; (ii) centre frequency, i.e. the frequency at which the magnitude of the increase or decrease in frequency response due to the bump is to be largest; and (iii) bandwidth (also known as ‘Q’, which is inversely proportional to bandwidth), i.e. the range of frequencies to be affected by the bump. Due to their low number of parameters, parametric equalizers can generally be quickly operated to achieve a desired effect on a particular audio source. However, parametric equalizers are generally not effective at achieving gentle audio effects across the whole audio band (or spectrum), nor are they effective at achieving a detailed but essentially arbitrary audio effect for room- or loudspeaker-correction.

[0005] Graphic (or graphical) equalizers use more processing than the above-mentioned tone controls and parametric equalizers. Graphic equalizers effectively divide the audio spectrum into a plurality (normally 31 or 61) of nominally independent, exponentially spaced, predetermined frequency bands. A gain control is then provided for each of these frequency bands. When the gains for these bands are controlled by vertical faders and these faders are arranged left to right in an increasing order of the frequencies they control, then the visual impression given is that the positions of the faders plot the frequency response produced by the graphic equalizer. Graphic equalizers are most frequently used to correct for problems in the final acoustic delivery of sound, such as problems in the loudspeaker, the room or both.

[0006] Graphic equalizers suffer from a number of problems that are mainly caused by the inevitable interaction between adjacent frequency bands. For example, if several adjacent faders are used to raise or lower the frequency response over a range of frequencies, then the resulting frequency response change is much greater than that actually indicated by the fader positions. This leads to excess effect over and above that intended by the user operating the faders. Furthermore, when several adjacent faders are used together in this way, the individual frequency bands do not combine smoothly to create a flat (either raised or lowered) frequency response. Instead, an unintended and subjectively annoying frequency response ripple is produced across these frequency bands. This makes it difficult to use a graphic equalizer to correct a room that presents problems over a broad range of frequencies.

[0007] Additionally (and at the opposite extreme to the above problems) it is difficult to correct or adjust the frequency response for a narrow band of frequencies using a graphic equalizer, since the individual filters which make up the graphic equalizer are at predetermined frequencies. These fixed frequencies cannot therefore be changed to the specific frequencies that the user wishes to correct.

[0008] Furthermore, due to the significantly increased amount of processing inherent to graphic equalizers when compared to tone controls and parametric equalizers (due to the combination of the outputs of a large number of filters), the phase response of a graphic equalizer is normally much more disruptive than is actually desirable.

[0009] It would therefore be desirable to provide improved audio equalization and filtering that addresses these problems.

[0010] According to a first aspect of the invention, there is provided a method of generating a desired shelving filter for audio data, the desired shelving filter having a low-corner frequency f₁ and a high-corner frequency f₂, wherein the difference between the gain of the frequency response of the desired shelving filter at f₂ and the gain of the frequency response of the desired shelving filter at f₁ is substantially equal to a non-zero desired amount, the method comprising: determining an order for a first shelving filter of a first predetermined type, wherein the low-corner frequency of the first shelving filter is f₁ and the high-corner frequency of the first shelving filter is f₂; determining, for a second shelving filter of a second predetermined type and of a predetermined order m, a frequency f₃ for the low-corner frequency of the second shelving filter and a frequency f₄ for the high-corner frequency of the second shelving filter, where f₃ is at least f₁, f₄ is greater than f₃, and f₄ is at most f₂; and forming the desired shelving filter as a filter combination of the first shelving filter and the second shelving filter; wherein f₃, f₄ and the order of the first shelving filter are determined such that difference between the gain of the frequency response of the filter combination at f₂ and the gain of the frequency response of the filter combination at f₁ is substantially equal to the desired amount.

[0011] This method allows a shelving filter with an arbitrary low-corner frequency and an arbitrary high-corner frequency, with an arbitrary difference in frequency response gain between the low- and high-corner frequencies to be specified and generated. This, in turn, will be a useful filter for generating an arbitrary desired frequency response, for example by combining several such shelving filters.

[0012] Some embodiments of the invention set m to be 1. This allows for a simpler implementation and a more straight-lined frequency response curve between the low- and high-corner frequencies.

[0013] In an embodiment of the invention, determining the order of the first shelving filter comprises determining the greatest order for the first shelving filter for which the magnitude of the desired amount is greater than the magnitude of the difference between the gain of the frequency response of the first shelving filter at f₂ and the gain of the frequency response of the first shelving filter at f₁. In an alternative embodiment of the invention, determining the order of the first shelving filter comprises determining the least order for the first shelving filter for which the magnitude of the desired amount is less than the magnitude of the difference between the gain of the frequency response of the first shelving filter at f₂ and the gain of the frequency response of the first shelving filter at f₁. These embodiments help ensure that a suitable second shelving filter can be determined.

[0014] In embodiments of the invention, if the desired amount is positive, then the first shelving filter is arranged such that the gain of the frequency response of the first shelving filter at f₂ is greater than the gain of the frequency response of the first shelving filter at f₁; and if the desired amount is negative, then the first shelving filter is arranged such that the gain of the frequency response of the first shelving filter at f₂ is less than the gain of the frequency response of the first shelving filter at f₁.

[0015] In an embodiment of the invention, determining f₃ and f₄ comprises: setting f₃ and f₄ relative to each other such that the difference between the gain of the frequency response of the second shelving filter at f₄ and the gain of the frequency response of the second shelving filter at f₃ is substantially equal to the difference between the desired amount and the difference between the gain of the frequency response of the first shelving filter at f₂ and the gain of the frequency response of the first shelving filter at f₁. This helps ensure that the desired shelving filter is generated when the first and the second shelving filters are combined to form the desired shelving filter.

[0016] To help avoid corner-effects (“wobbles” and unwanted curves) at and around the low- and high-corner frequencies, some embodiments of the invention set f₃ and f₄ so that the range of frequencies from f₃ to f₄ is substantially centred in the range of frequencies from f₁ to f₂. In particular, f₃ and f₄ may be set so that f₄+f₃=f₂+f₁ or f₄f₃=f₂f₁.

[0017] In an embodiment in which m=1, if the desired amount is positive then: each zero of the transfer function of the first shelving filter is located at a respective location determined by interpolating between (a) the location of a corresponding pole of the transfer function of a first low-pass filter of a type corresponding to the first predetermined type and having a cut-off frequency of f₁ and an order of n−1 and (b) the location of a corresponding pole of the transfer function of a second low-pass filter of that corresponding type and having a cut-off frequency of f₁ and an order of n; and each pole of the transfer function of the first shelving filter is located at a respective location determined by interpolating between (c) the location of a corresponding pole of the transfer function of a third low-pass filter of that corresponding type and having a cut-off frequency of f₂ and an order of n−1 and (d) the location of a corresponding pole of the transfer function for a fourth low-pass filter of that corresponding type and having a cut-off frequency of f₂ and an order of n. A shelving filter formed in this manner is particularly useful when it is desired to be able to transition from one shelving filter into another shelving filter.

[0018] In an embodiment in which m=1, if the desired amount is negative then: each pole of the transfer function of the first shelving filter is located at a respective location determined by interpolating between (a) the location of a corresponding pole of the transfer function of a first low-pass filter of a type corresponding to the first predetermined type and having a cut-off frequency of f₁ and an order of n−1 and (b) the location of a corresponding pole of the transfer function of a second low-pass filter of that corresponding type and having a cut-off frequency of f₁ and an order of n; and each zero of the transfer function of the first shelving filter is located at a respective location determined by interpolating between (c) the location of a corresponding pole of the transfer function of a third low-pass filter of that corresponding type and having a cut-off frequency of f₂ and an order of n−1 and (d) the location of a corresponding pole of the transfer function for a fourth low-pass filter of that corresponding type and having a cut-off frequency of f₂ and an order of n.

[0019] In some embodiments of the invention, a user may specify f₁, f₂ and the desired amount, thereby making the shelving filter fully configurable by the user.

[0020] In some embodiments of the invention, the first predetermined type and the second predetermined type are each filter types based on properties of a respective one of: Butterworth low-pass filters; Chebychev low-pass filters; and elliptical low-pass filters. These properties may include the positions of zeros and poles of transfer functions for low-pass filters of the corresponding filter type. The first predetermined type may be the same as the second predetermined type.

[0021] According to a second aspect of the invention, there is provided a method of generating a desired filter for audio data, the method comprising: specifying a plurality of frequency response indicators to define a desired frequency response for the desired filter, each frequency response indicator specifying a user-defined frequency and a corresponding user-defined gain for the desired frequency response at that frequency; and determining a set of one or more filters such that, for each of the frequency response indicators, the gain of the frequency response of a combination of the one or more filters in the set of filters at the frequency specified by that frequency response indicator is substantially equal to the gain specified by that frequency response indicator.

[0022] With such a method, a very flexible audio equalizer can be generated according to a user's specific requirements. In particular, the user is not limited to predefined frequency bands as in the traditional graphics equalizer, nor is the user limited to the number of such frequency bands as some embodiments allow the number of frequency response indicators to be user-defined.

[0023] Some embodiments of the invention comprise the user defining one or more of the plurality of frequency response indicators.

[0024] In an embodiment of the invention, determining the set of one or more filters comprises: for each pair of adjacent frequency response indicators, generating a corresponding shelving filter, wherein a frequency response indicator specifying a frequency f_A and another frequency response indicator specifying a frequency f_B greater than f_A are adjacent if none of the plurality of frequency response indicators specifies a frequency between f_A and f_B.

[0025] In particular, the method may comprise: for each frequency response indicator, maintaining a corresponding target gain; in which, for a pair of adjacent frequency response indicators comprising a first frequency response indicator specifying a frequency f_A and an adjacent second frequency response indicator specifying a frequency f_B greater than f_A, the step of generating a shelving filter comprises carrying out a method according to the above first aspect of the invention, in which: f₁ is based on f_A; f₂ is based on f_B; and the desired amount is set to be the difference between the target gain corresponding to the second frequency response indicator and the target gain corresponding to the first frequency response indicator. This may involve initialising each target gain to the gain specified by the corresponding frequency response indicator.

[0026] In such embodiments, the method may comprise, for a pair of adjacent frequency response indicators comprising a first frequency response indicator and an adjacent second frequency response indicator: before generating the shelving filter corresponding to that pair of adjacent frequency response indicators, determining whether the order of that shelving filter will exceed a predetermined maximum threshold and, if so, adjusting one or both of the target gain corresponding to the first frequency response indicator and the target, gain corresponding to the second frequency response indicator to reduce the order of that shelving filter. This helps ensure that the order of the shelving filters that are generated are not too large, thereby avoiding excessive power consumption, hardware usage, processing time and delay.

[0027] In some of these embodiments, for a pair of adjacent frequency response indicators comprising a first frequency response indicator specifying a frequency f_A and an adjacent second frequency response indicator specifying a frequency f_B greater than f_A, the step of carrying out a method according to the first aspect of the invention to generate the shelving filter corresponding to that pair of adjacent frequency response indicators comprises setting f₁ equal to f_A and setting f₂ equal to f_B.

[0028] Alternatively, in some of these embodiments, the method comprises: for each pair of adjacent frequency response indicators, maintaining a corresponding frequency-proportion value; and, for a pair of adjacent frequency response indicators comprising a first frequency response indicator specifying a frequency f_A and an adjacent second frequency response indicator specifying a frequency f_B greater than f_A, the step of carrying out a method according to the above first aspect of the invention to generate the shelving filter corresponding to that pair of adjacent frequency response indicators comprises determining f₁ and f₂ based on f_A, f_B and the corresponding frequency-proportion value. In particular, f₁ and f₂ may be determined so that f₂−f₁=h(f_B−f_A) or f₂/f₁=(f_B/f_A)^h, where h is the frequency-proportion value corresponding to that pair of adjacent frequency response indicators.

[0029] Then, the range from f₁ to f₂ may be substantially centred relative to the range from f_A to f_B, for example, f₁ and f₂ may be determined so that f₂+f₁=f_B+f_A or f₂f₁=f_Af_B.

[0030] Some embodiments of the invention may comprise, for a pair of adjacent frequency response indicators comprising a first frequency response indicator and an adjacent second frequency response indicator: before generating the shelving filter corresponding to that pair of adjacent frequency response indicators, determining whether the order of that shelving filter will exceed a predetermined maximum threshold and, if so, increasing the frequency-proportion value corresponding to that pair of adjacent frequency response indicators to reduce the order of that shelving filter. Some of these embodiments may be arranged such that the frequency-proportion value may not be increased above a predetermined maximum frequency-proportion value.

[0031] Each frequency proportion value may be initialised to a corresponding initialisation value, which is preferably in the range from about ⅓ to about ½.

[0032] Some embodiments of the invention may then comprise: after generating the set of one or more filters, performing one or more iterations, wherein an iteration comprises: adjusting one or more of the target gains; and repeating the step of generating using the adjusted target gains to update the set of one or more filters; wherein the step of adjusting comprises adjusting one or more of the target gains so that an error between the desired frequency response and the frequency response of a combination of the one or more filters in the updated set of filters is less than an error between the desired frequency response and the frequency response of a combination of the one or more filters in the set of filters prior being updated.

[0033] Such an embodiment may involve performing at most a predetermined maximum number of iterations. This ensures that the desired filter is generated within a limited period of time, which is particularly useful for real-time audio processing applications.

[0034] Each of these iterations may comprise determining an error between the desired frequency response and the frequency response of a combination of the one or more filters in the set of filters, wherein the step of performing one or more iterations is terminated if the determined error is less than a predetermined error threshold. Again, this ensures that the desired filter is generated within a limited period of time (in this case, when the generated set of filters has a frequency response sufficiently close to the desired frequency response) which is particularly useful for real-time audio processing applications. Determining an error may comprise: for each of the frequency response indicators, determining the difference between the gain of the desired frequency response and the gain of the frequency response of the combination of the one or more filters in the set of filters at the frequency specified by that frequency response indicator; and determining the error based on the determined differences.

[0035] In some embodiments of the invention, adjusting a target gain corresponding to a frequency response indicator comprises subtracting, from that target gain, the difference between the gain of the frequency response of the combination of the filters in the set of filters at the frequency specified by that frequency response indicator and the gain specified by that frequency response indicator.

[0036] Some embodiments of the invention comprise: a user specifying a gain scale factor; and for each of the frequency response indicators, scaling the gain specified by that frequency response indicator by the specified gain scale factor. This provides the user with a mechanism for easily controlling the overall intensity of the desired frequency response that the user has specified via the frequency response indicators.

[0037] According to a third aspect of the invention, there is provided a method of implementing an audio filter, in which the transfer function of the audio filter comprises 2s complex-valued zeros and 2s complex-valued poles, the method comprising: forming s second-order-filter-sections, such that, for each integer i in the range 1≦i≦s, the i-th second-order-filter-section is based on: (a) the complex-valued zero of the transfer function of the audio filter that, together with the associated complex conjugate zero, is the i-th closest pair of complex-valued zero and associated complex conjugate zero of the transfer function of the audio filter to the unit circle; and (b) the complex-valued pole of the transfer function of the audio filter that, together with the associated complex conjugate pole, is the i-th closest pair of complex-valued pole and associated complex conjugate pole of the transfer function of the audio filter to the unit circle.

[0038] According to a fourth aspect of the invention, there is provided a method of changing from filtering audio data with a first audio filter to filtering the audio data with a second audio filter, in which the transfer function of the first audio filter comprises 2s complex-valued zeros and 2s complex-valued poles, and in which the transfer function of the second audio filter comprises 2t complex-valued zeros and 2t complex-valued poles, the method comprising: forming the first filter using second-order-filter-sections, in which, for each integer i in the range 1≦i≦s, the i-th second-order-filter-section is based on (a) the complex-valued zero of the transfer function of the first audio filter that, together with the associated complex conjugate zero, is the i-th closest pair of complex-valued zero and associated complex conjugate zero of the transfer function of the first audio filter to the unit circle; and (b) the complex-valued pole of the transfer function of the first audio filter that, together with the associated complex conjugate pole, is the i-th closest pair of complex-valued pole and associated complex conjugate pole of the transfer function of the first audio filter to the unit circle; forming the second filter using t second-order-filter-sections, in which, for each integer j in the range 1≦j≦t, the j-th second-order-filter-section is based on (c) the complex-valued zero of the transfer function of the second audio filter that, together with the associated complex conjugate zero, is the j-th closest pair of complex-valued zero and associated complex conjugate zero of the transfer function of the second audio filter to the unit circle; and (d) the complex-valued pole of the transfer function of the second audio filter that, together with the associated complex conjugate pole, is the j-th closest pair of complex-valued pole and associated complex conjugate pole of the transfer function of the second audio filter to the unit circle; and for each integer k in the range 1≦k≦minimum(s,t), setting the value of a state variable for the k-th second-order-filter-section of the second audio filter to be the value of a corresponding state-variable for the k-th second-order-filter-section of the first audio filter.

[0039] The use of audio filters formed according to the third aspect of the invention allows the method of the fourth aspect of the invention to be used. Such an arrangement for transitioning from one filter to another helps minimize the amount of audio noise that may be introduced during the transitional phase between the filters.

[0040] In one embodiment of the invention, if t is greater than s, then the method comprises, for each integer r in the range s+1≦r≦t: initialising the state variables for the r-th second-order-filter section of the second audio filter to the value 0; providing audio input to the r-th second-order-filter section of the second audio filter for a predetermined period of time after beginning to change from filtering the audio data with the first audio filter to filtering the audio data with the second audio filter; and after the predetermined period of time, starting to use the r-th second-order-filter section of the second audio filter to filter the audio data.

[0041] According to a fifth aspect of the invention, there is provided a method of processing audio data, the method comprising: filtering the audio data by applying a first desired filter to the audio data, the first desired filter being a filter generated by a method according to the above-mentioned methods for generating a desired filter. This may comprise changing the filtering of the audio data from applying the first desired filter to the audio data to applying a second desired filter to the audio data, the second desired filter being a filter generated by a method according to the above-mentioned methods for generating a filter.

[0042] When the first and second desired filters are generated by a method based on specifying frequency response indicators, the method may comprise: specifying a plurality of frequency response indicators to generate the first desired filter; updating the plurality of frequency response indicators; and generating the second desired filter based on the updated plurality of frequency response indicators.

[0043] In some embodiments of the invention, the step of changing comprises carrying out a method according the above-mentioned fourth aspect of the invention for changing from filtering audio data with a first audio filter to filtering the audio data with a second audio filter.

[0044] According to a sixth aspect of the invention, there is provided a method of configuring a processor for processing audio data, the method comprising: generating a desired filter for the audio data by carrying out a method according to any one of the above-mentioned methods of generating a filter; and configuring the processor to process the audio data using the generated filter.

[0045] According to a seventh aspect of the invention, there is provided an apparatus for generating a desired audio filter, the apparatus comprising a processor arranged to generate a desired filter for audio data by carrying out a method according to any one of the above-mentioned methods of generating a filter.

[0046] According to an eighth aspect of the invention, there is provided an apparatus for processing audio data, the apparatus comprising: a processor arranged to process audio data by carrying out one of the above-mentioned methods for processing audio data; and means for supplying audio data to the processor for processing. This processor may be arranged to generate a desired filter for processing the audio data by carrying out any of the above-mentioned methods for generating a desired filter.

[0047] According to a ninth aspect of the invention, there is provided an audio filter generated by any one of the above-mentioned methods for generating a desired audio filter.

[0048] According to a tenth aspect of the invention, there is provided a computer program which, when executed by a computer, carries out a method according to any one of the above-mentioned methods. The computer program may be carried on a data carrying medium, which may be a storage medium or a transmission medium.

[0067] In the description that follows and in the figures, certain embodiments of the invention are described. However, it will be appreciated that the invention is not limited to the embodiments that are described and that some embodiments may not include all of the features that are described below. It will be evident, however, that various modifications and changes may be made herein without departing from the broader scope of the invention as set forth in the appended claims.

[0068] FIG. 12 schematically illustrates a method of generating a desired shelving filter 2010 for audio data. The desired shelving filter 2010 has a low-corner frequency f₁ and a high-corner frequency f₂. The difference between the gain of the frequency response of the desired shelving filter at f₂ and the gain of the frequency response of the desired shelving filter at f₁ is substantially equal to a non-zero desired amount.

[0069] An order for a first shelving filter 2015 of a first predetermined type is determined 2035. The low-corner frequency of the first shelving filter is f₁ and the high-corner frequency of the first shelving filter is f₂.

[0070] A frequency f₃ for the low-corner frequency of a second shelving filter 2020 of a second predetermined type and of a predetermined order m is determined 2035. A frequency f₄ for the high-corner frequency of the second shelving filter is determined 2035 where f₃ is at least f₁, f₄ is greater than f₃ and f₄ is at most f₂.

[0071] The desired shelving filter 2010 is formed 2025 as a filter combination 2030 of the first shelving filter 2015 and the second shelving filter 2020. Here, f₃, f₄ and the order of the first shelving filter 2015 are determined 2035 such that difference between the gain of the frequency response of the filter combination 2030 at f₂ and the gain of the frequency response of the filter combination 2030 at f₁ is substantiall egual to the desired amount.

(1) System Overview

[0072] FIG. 1 schematically illustrates a system 100 according to an embodiment of the invention. The system 100 comprises a device 102 that itself comprises an audio interface 104, a data interface 106, a user input interface 108, a memory 110, a graphical interface 112 and a processor 114. The device 102 may be any type of device for carrying out audio processing. For example, the device 102 may be a standard personal computer. Alternatively, the device 102 may be a dedicated audio processing apparatus.

[0073] The processor 114 is arranged to carry out a method according to an embodiment of the invention (such methods being described in more detail later). To achieve this, the processor 114 may comprise a hardware device, such as one or more field-programmable-gate-arrays (FPGAs) and/or application-specific-integrated-circuits (ASICs) that have been produced and configured specifically for carrying out such a method. The processor 114 may be a digital-signal-processor (DSP) chip that has been configured specifically for carrying out a method according to an embodiment of the invention.

[0074] Alternatively, to carry out a method according to an embodiment of the invention, the processor 114 may be a more general purpose processor that is capable of executing a computer program that itself comprises program instructions for performing such a method. The computer program may be a specific self-contained executable application for carrying out an embodiment of the invention. Alternatively, the computer program may be configured as a plug-in (such as a dynamic-link-library, or DLL) for a more general purpose audio processing software application, with the plug-in adding functionality according to an embodiment of the invention to that general purpose audio processing software application. Methods for generating and using plug-ins are well-known.

[0075] The memory 110 may be any type of memory for storing data used by the processor 114 when carrying out an embodiment of the invention. In some embodiments, the memory 110 forms part of the processor 114 itself. The memory 110 may be a read-only-memory, with the processor 114 then being configured to be able to read data from the memory 110 but not to write data to the memory 110. Alternatively, the memory 110 may be a random-access-memory, with the processor 114 being configured to be able to both read data from and write data to the memory 110.

[0076] The memory 110 may store the whole or part of a computer program that comprises program instructions for carrying out a method according to an embodiment of the invention, with the processor 114 then being arranged to read and execute this computer program to carry out an embodiment of the invention. The memory 110 may store configuration data (for example information regarding a desired frequency response for an audio filter, as described later). The memory 110 may store audio data to be processed (filtered or equalized) by the processor 114. The processor 114 may also use the memory 110 to store the output of its processing (such as filter coefficients for filters that it has generated or audio data that it has processed) and/or intermediate data and results from its processing.

[0077] The data interface 106 may be any interface through which the processor 114 may receive and/or output/transmit data. For example, the data interface 106 may comprise one or more of: a magnetic medium reader and/or writer for reading data from and/or writing data to a magnetic storage medium (such as a hard disk drive); an optical medium reader and/or writer for reading data from and/or writing data to an optical storage medium, such as a CD-ROM, a DVD-ROM or a BluRay disk; a solid-state memory device interface for reading data from and/or writing data to a solid-state memory device (such as a flash-memory device); and a network interface for receiving data from and/or transmitting data to a computer over a network (such as the Internet, a local area network, a wide area network, a metropolitan area network, etc.). In this way, the processor 114 may receive digital data from one or more sources and can output digital data to one or more destinations. Data received via the data interface 106 may be stored directly in the memory 110, or may be passed first to the processor 114 for processing and/or subsequent storage in the memory 110. Additionally, the data interface 106 may receive data for outputting or for storage on a storage medium either directly from the memory 110 or from the processor 114.

[0078] If the processor 114 uses a computer program that comprises program instructions for carrying out a method according to an embodiment of the invention, then the processor 114 may receive the computer program from a source via the data interface 106. Configuration data (for example information regarding a desired frequency response for an audio filter, as described later) or commands to perform processing may also be received via the data interface 106. Furthermore, audio data to be processed (filtered or equalized) by the processor 114 may be received via the data interface 106 (for example as one or more audio data files or data streams). The processor 114 may also output the results of its processing (such as filter coefficients that it has generated or audio data files or audio data streams that it has produced) for storage or transmission via the data interface 106.

[0079] The processor 114 is also arranged to receive audio data from the audio interface 104. The audio interface 104 may be any interface arranged to receive audio signals (such as live or pre-recorded audio signals, for example: audio signals from one or more microphones and/or one or more audio reproduction devices—not shown in FIG. 1). Some of these audio signals may be received in digital form. Alternatively, some of these audio signals may be received in analog form, in which case, if the processor 114 is arranged to perform digital processing then the audio interface 104 may be arranged to perform analog-to-digital conversion to generate corresponding digital audio data. The audio interface 104 may be arranged to simply pass the input audio signals to the processor 114 for processing, or it may be arranged to write audio data to the memory 110 for storage so that the processor 114 can subsequently process the stored audio data. Additionally, processed audio data may be output via the audio interface 104 (for example output to a speaker or to an audio recording device—not shown in FIG. 1) either from the memory 110 or from the processor 114 itself. In this case, the audio interface 104 may be arranged to perform digital-to-analog conversion in order to output digital audio data as an analog audio signal, for example to a speaker.

[0080] The processor 114 may perform its processing in accordance with input received from a user via the user interface 108. The user interface 108 receives input signals from one or more user input devices, such as a keyboard 116 and a mouse 118, although it will be appreciated that other forms of input device may be used as are well-known in this field of technology. The user interface 108 then informs the processor 114 of these input signals so that the processor 114 may then respond to the input from the user accordingly.

[0081] The processor 114 also communicates with the graphical interface 112, which controls a display 120 (such as a screen or a monitor). The processor 114 is arranged to instruct the graphical interface 112 to output a visual display via the display 120. This will be described in more detail with reference to FIG. 3 below. The graphical interface 112 then outputs the visual display accordingly. Additionally, the display 120 may be touch-sensitive or may provide other means through which the user can provide input to the processor 114, so that the processor 114 can then receive commands and/or configuration data via the display 120 as well as, or as an alternative to, via the user interface 108.

[0082] It will be appreciated that the system 100 need not comprise all of the elements mentioned above, and that various elements of the system 100 may be provided or omitted according to a specific desired configuration of the system. For example, if audio data is to be received via only via the data interface 106, then the system 100 may omit the audio interface 104, and vice versa. As a further example, if configuration data (such as a desired frequency response for the processing) is to be received via a user input via the user interface 108, then the system 100 may be configured to not receive such data via the data interface 106 or via the display 120.

[0083] Additionally, it will be appreciated that the system 100 may be formed by a single device 102, or the system 100 may be formed from multiple separate devices linked together, each contributing to one or more of the features of the system 100. For example:

[0084] (i) A first system 100a may be formed from a dedicated audio processing device. This device may comprise the audio interface 104a, the memory 110a and a processor 114a (for example, in the form of a DSP chip). The processor 114a may be configured to process audio data received via the audio interface 104a and output processed audio via the audio interface 104a. The processing of the audio data may involve filtering the data in accordance with an embodiment of the invention (as described later). The first system 100a could be a system installed, for example, in a stadium, a concert hall, etc. for processing sound at such a venue. The system 100a may be lacking means or functionality to allow a user to vary the configuration of the audio processing that it performs.

[0085] (ii) A second system 100b may be formed from a general purpose computer. The computer may be arranged to carry out an embodiment of the invention for determining a filter and determining filter coefficients in accordance with a user's requirements (as explained in more detail later). However, the second system 100b may, or may not, be configured to perform processing of audio data in accordance with the filters and filter coefficients that it has generated.

[0086] (iii) A third system 100c may be formed by coupling the above first system 100a and the above second system 100b. In this way, the second system 100b may be used to generate filter coefficients and then configure the first system 100a with those generated filter coefficients. Thus, for example, an audio engineer may visit a venue at which the first system 100a is installed for processing audio at that venue. The audio engineer may bring along the second system 100b (for example a laptop computer) and, using the second system 100b, configure the first system 100a accordingly. Once this has been done, the second system 100b can be decoupled from the first system 100a, leaving the first system 100a to process audio according to the configuration it has been provided with. Thus, in this third system 100c, various components (such as the memory 110 and the processor 114) may be formed from multiple separate components (such as the respective memories and processors of the first and second systems).

(2) Overview of the Processing of the System of FIG. 1

[0087] FIG. 2 is a flowchart schematically illustrating an overview of the processing 200 performed by the processor 114 in accordance with an embodiment of the invention.

[0088] The processing 200 makes use of a set of one or more frequency response indicators. A frequency response indicator is data specifying (i) a user-defined frequency and (ii) a corresponding used-defined desired gain (positive, negative or zero) for a desired frequency response at that frequency. In this way, a desired frequency response may be indicated (or defined or specified), i.e. the frequency response indicators define a desired frequency response for subsequent audio processing.

[0089] At a step S202, the set of frequency response indicators is initialised. This could be, for example, setting there to be no frequency response indicators so that the set of frequency response indicators is initially empty. Alternatively, the processor may have been provided with configuration data (in a manner as discussed above, for example from a storage medium or from a remote computer over a network) that specifies one or more initial frequency response indicators. A graphical user interface to allow a user to specify frequency response indicators will be discussed later with respect to FIG. 3.

[0090] Then, at a step S204, a set of one or more filters is determined with the aim that, for each of the frequency response indicators in the set of frequency response indicators, the gain of the frequency response resulting from the combination of the filters in the determined set of filters at the frequency specified by that frequency response indicator is substantially the same as the corresponding desired gain specified by that frequency response indicator. Methods for determining the set of filters will be described later.

[0091] The processor 114 may then determine the filter coefficients corresponding to the filters in the determined set of filters. The processor 114 may store these coefficients in the memory 110 (for example, for use in a step S206 described below) and/or may output these coefficients via the data interface 106 for storage on a storage medium and/or transmission to another computer.

[0092] The combination of the set of filters generated at the step S204 is then an infinite-impulse-response filter suitable for applying to audio data.

[0093] The processor 114 may be arranged to process audio data using the set of filters generated at the step S204 during the processing 200 of FIG. 2, so that a user can set and update the set of frequency response indicators to listen to the effect on audio data. This may be used, for example, when experimenting with the frequency response indicators to generate a desired effect, or when processing a live audio feed. Thus, at an optional step S206, the processor 114 applies the set of filters generated at the step S204 to audio data. Methods for applying filters to audio data are well-known and shall not be described in more detail herein. As mentioned, the audio data to which the filters are applied may already be stored in the memory 110 or may be received via the audio interface 104 or via the data interface 106. The processed audio data may then be stored in the memory 110. The processed audio data may be output and/or stored via the audio interface 104 and/or via the data interface 106.

[0094] At a step S208, the set of frequency response indicators is updated so as to specify a new set of frequency response indicators. This may be performed, for example, by a user. The update may involve: (i) changing the number of frequency response indicators (for example by adding one or more new frequency response indicators and/or deleting one or more existing frequency response indicators); and/or (ii) changing the frequency value corresponding to one or more of the frequency response indicators; and/or (iii) changing the gain value corresponding to one or more of the frequency response indicators. Again, methods for doing this will be described in more detail later.

[0095] Processing then returns to the step S204 at which a set of filters is generated based on the updated set of frequency response indicators. Then, at the step S206, the processor 114 transitions from applying the previous set of filters to applying the newly generated set of filters to the audio data. Methods for carrying out this transition will be described in more detail later.

(3) Graphical User Interface

[0096] FIG. 3 schematically illustrates an example graphical user interface (GUI) 300 provided by the processor 114 to a user on the display 120 according to an embodiment of the invention. This GUI 300 may be used to specify frequency response indicators at the step S202 of FIG. 2 and to update the frequency response indicators at the step S208 of FIG. 2.

[0097] It will be appreciated that other embodiments of the invention may omit some of the features displayed on the GUI 300 shown in FIG. 3 and/or may include additional features on the GUI 300.

[0098] The GUI 300 displays a graph 301, or a plot, of the frequency response of the set of filters generated at the step S204 of FIG. 2. The horizontal axis of the graph 301 represents log-frequency and the vertical axis of the graph 301 represents the gain (or size or magnitude of the change in the power of an audio signal due to the application of the generated set of filters), measured in decibels. It will be appreciated, though, that other units of measurement could be used instead.

[0099] The GUI 300 allows the user to define, or specify, a set of frequency response indicators 302. Each frequency response indicator 302 specifies a user-defined frequency and a corresponding used-defined desired gain for a desired frequency response at that frequency. In other words, each frequency response indicator 302 identifies a point on the graph 301 through which the user wishes a desired frequency response to pass. The frequency response indicators 302 are displayed as handles, or pointers, on the GUI 300.

[0100] The GUI 300 allows the user to add or remove frequency response indicators 302. Thus, the number of frequency response indicators 302 to be used may be determined by the user. Additionally, the GUI 300 allows the user to move an existing frequency response indicator 302 along both the horizontal (log-frequency) axis and the vertical (gain) axis. Thus, both the frequency and the gain associated with a frequency response indicator 302 can be set by a user. In particular, the frequency for a frequency response indicator 302 is not limited to frequencies from a predetermined finite set of frequencies.

[0101] The user may add, adjust or delete a frequency response indicator 302 in a number of ways. For example, the user may use the keyboard 116 and/or the mouse 118 and/or a touch sensitive display 120 to select a location on the graph 301 so that a new frequency response indicator 302 is generated that specifies the corresponding frequency and gain of the selected location on the graph 301. The user may use the keyboard 116 and/or the mouse 118 and/or a touch sensitive display 120 to select an existing frequency response indicator 302 and then move/drag that frequency response indicator 302 to a new location on the graph 301 to adjust that selected frequency response indicator 302 accordingly.

[0102] Alternatively, the user may use a frequency input box 306 and a gain input box 308 to specify (by inputting numerical values) a frequency and a desired gain respectively. Thus, the resolution of the display 120 does not limit the accuracy with which the frequency response indicators 302 may be set by the user. A button 310 may then be provided on the GUI 300 which, when selected by the user, generates a new frequency response indicator 302 based on the frequency and gain entered at the frequency input box 306 and the gain input box 308. When the user selects a frequency response indicator 302, the GUI 300 may cause the frequency input box 306 and the gain input box 308 to display the corresponding frequency and gain for that selected frequency response indicator 302. The user may then change the displayed frequency and/or gain and then use a button 312 provided on the GUI 300 which sets (or adjusts) that selected frequency response indicator 302 so that its frequency and gain are those entered at the frequency input box 306 and the gain input box 308.

[0103] A frequency response indicator 302 may be deleted by the user selecting that frequency response indicator 302 using the keyboard 116 and/or the mouse 118 and/or a touch sensitive display 120, and then pressing a delete key on the keyboard 116 or selecting a delete button 314 provided on the GUI 300.

[0104] The GUI 300 may allow the user to select multiple frequency response indicators 302 so as to group those frequency response indicators 302 together. In this way, this selected group of frequency response indicators 302 may then be deleted together or adjusted together (for example, by moving them all as a group, either horizontally to change their frequencies and/or vertically to change their associated gains). This allows for easier manipulation of the frequency response indicators 302 by an operator of the system 100.

[0105] The user sets the frequency response indicators 302 (and the number of frequency response indicators 302) to define the desired frequency response for audio processing (filtering or equalisation). As mentioned, the processor 114 is arranged to generate (at the step S204) a set of filters based on the set of frequency response indicators 302. This will be described in more detail below. The processor 114 may then determine the frequency response resulting from this generated set of filters and plot this on the graph 301 as a curve 304. Ideally, this curve 304 will match the frequency response desired by the user and will thus pass through the points representing the frequency response indicators 302. However, as described in more detail later, the calculated frequency response curve 304 may miss one or more of the points representing the frequency response indicators 302.

[0106] In some embodiments of the invention, the set of filters is generated at the step S204 each time the user adjusts, adds or deletes a frequency response indicator 302. In this way, the user is continuously provided with the most up-to-date curve 304 showing the frequency response corresponding to the generated set of filters for the currently specified frequency response indicators 302. Additionally, if the processor 114 is processing audio data based on the generated set of filters, then this approach allows the set of filters used for the audio processing to be varied in accordance with each modification that the user makes to the set of frequency response indicators 302. In an alternative embodiment, though, the processor 114 may display an updated frequency response curve 304 to the user 104, but may only start to use the updated set of filters at the step S206 once the user has selected an apply button 316 provided on the GUI 300. In this way, the user can visually see the frequency response curve 304 and modify the frequency response indicators 302 until a desired curve 304 is achieved, and then apply the corresponding set of filters to the audio data without having to have applied intermediate sets of filters (which may have been audibly undesirable) to the audio data.

[0107] Alternatively, the processor 114 may be arranged to only generate a new set of filters at the step S204 (i.e. move from the step S202 to the step S204) once the user is content with the set of frequency response indicators that he has specified and/or updated and has then selected the apply button 316. In this embodiment, selecting the apply button 316 then causes the processor 114 to determine, or generate, a corresponding set of filters at the step S204 and then display the corresponding frequency response curve 304.

[0108] It will be appreciated that embodiments of the invention may make use of other methods for specifying and receiving the frequency response indicators. Thus, some embodiment may use the GUI 300 as an additional method for specifying and updating the frequency response indicators, whilst other embodiments may not use the GUI 300 at all. For example, frequency response indicators may be saved as configuration data in the memory 110, on a storage medium accessible by the processor 114 via the data interface 106, or at a remote computer accessible via the data interface 106. This configuration data may specify just one set of frequency response indicators. Alternatively, the configuration data may specify an ordered plurality of sets of frequency response indicators, with corresponding timing information specifying when one set of frequency response indicators is to be updated, to a subsequent (or next) set of frequency response indicators. In this way, the set of frequency response indicators may be automatically updated by the processor 114 whilst processing audio data. The GUI 300 therefore may provide a load button 318 to allow the user to locate and access a source of such frequency response indicator configuration data.

[0109] Additionally, the GUI 300 may provide a save button 320 to allow a user to save, or output, the current set of frequency response indicators either to the memory 110, or to a storage medium or a remote computer via the data interface 106. Additionally, or alternatively, the save button 320 may allow the user to save, or output, the coefficients of the filters that have been generated at the step S204 for the current set of frequency response indicators 302 either to the memory 110, or to a storage medium or a remote computer via the data interface 106.

(4) Shelving Filters

[0110] A shelving filter (or shelf filter) is a filter whose frequency response has a gain that: (i) is substantially equal to a first gain value g₁ for frequencies below a first frequency f₁; (ii) is substantially equal to a second gain value g₂ (different from the first gain value g₁) for frequencies above a second frequency f₂ (where f₁<f₂); and (iii) changes continuously (and usually monotonically and, even more usually, strictly monotonically) from g₁ to g₂ as the frequency changes from f₁ to f₂.

[0111] In this description, f₁ shall be referred to as the “low-corner frequency” and f₂ shall be referred to as the “high-corner frequency” (the words “low” and “high” here indicating that f₁ is less than f₂, and not implying that g₁ is less than g₂). Frequency f₁ may also be referred to as the “lower-corner frequency” and frequency f₂ may also be referred to as the “upper-corner frequency”.

[0112] If g₁>g₂, then the shelving filter is a so-called “sinking shelving filter” or a “falling shelving filter”. If g₁<g₂, then the shelving filter is a so-called “rising shelving filter”. Of course, g₁ may be positive, negative or zero-valued. Similarly, g₂ may be positive, negative or zero-valued, independently of g₁.

[0113] FIG. 4a illustrates the frequency response graph for an example falling shelving filter; FIG. 4b illustrates the frequency response graph for an example rising shelving filter. In this application, all frequency response graphs (including those of FIGS. 4a and 4b) have a horizontal axis representing log-frequency and a vertical axis representing gain measured in decibels (dB).

[0114] As mentioned, the frequency response of the shelving filter has a gain that is substantially equal to g₁ for frequencies below the low-corner frequency f₁. However, it will be appreciated that the frequency response gain need not be exactly equal to g₁ for all frequencies below f₁, but may instead simply be different from g₁ by an amount that is small in comparison to the difference between g₁ and g₂. Similarly, whilst the frequency response of the shelving filter has a gain that is substantially equal to g₂ for frequencies above the high-corner frequency f₂, it will be appreciated that the frequency response need not be exactly equal to g₂ for all frequencies above f₂, but may instead simply be different from g₂ by an amount that is small in comparison to the difference between g₁ and g₂. Thus, the frequency response below f₁ and above f₂ need not be perfectly flat. In general, though, the gain of the frequency response tends asymptotically to g₁ as the frequency decreases below f₁, and the gain of the frequency response tends asymptotically to g₂ as the frequency increases above f₂. FIG. 4c therefore illustrates the frequency response graph for an example falling shelving filter based on FIG. 4a, but showing how the frequency response may vary for frequencies below the low-corner frequency f₁ and for frequencies above the high-corner frequency f₂. Similarly, FIG. 4d illustrates the frequency response graph for an example rising shelving filter based on FIG. 4b, but showing how the frequency response may vary for frequencies below the low-corner frequency f₁ and for frequencies above the high-corner frequency f₂.

[0115] As is known in this field of technology, the order, n, of a filter is the term used in this field of technology to indicate the polynomial order of the numerator and of the denominator of the transfer function for that filter. Thus, the transfer function for an n-th order filter has n zeros and n poles. The order n is, naturally, a non-negative integer.

[0116] Described below are a number of methods for generating a shelving filter. Except where stated otherwise, these methods are based on properties of low-pass filters of a predetermined type. The predetermined type may be, for example, Butterworth low-pass filters, elliptical low-pass filters, Chebychev low-pass filters, etc. As these types of low-pass filter are well-known in this field of technology, they shall not be described in more detail herein. The predetermined type of low-pass filter whose properties are used will be referred to as the “basis-type” of low-pass filter (as such low-pass filters form the basis for the shelving filters that are formed). Thus, the basis-type may be, for example: “Butterworth”, “elliptical”, “Chebychev”, etc. Low-pass filters of the basis-type will be referred to as “low-pass basis-filters”. The methods described below apply equally to these various different basis-types. Hence, the rest of the description will use Butterworth filters as the basis-type of filter, but it will be appreciated that the methods described can be applied equally using other basis-types too.

[0117] As is also known, for an n-th order low-pass basis-filter of basis-type B, the frequency response has a rate-of-change of gain that is substantially −r_B×n dB per octave at frequencies above the filter's cut-off frequency, where r_B is a positive value dependent upon the basis-type B. Thus, for a frequency f above the cut-off frequency, the frequency response of the n-th order low-pass basis-filter at the frequency 2f has a gain that is substantially r_B×n dB less than at frequency f, i.e. the gradient of the frequency response curve plotted on a log-frequency vs gain graph is substantially −r_B×n for frequencies above the cut-off frequency. For Butterworth filters, r_Butterworth=20 log₁₀(2)≈6.0 dB per octave.

(4.1) Method A

[0118] An example method for designing an n-th order shelving filter is given in “Parametric Recursive Higher-Order Shelving Filters”, Martin Holters and Udo Zölzer, Audio Engineering Society, May 2006. This paper describes that an n-th order shelving filter may be based on n-th order low-pass Butterworth filters as discussed below. It also mentions that Chebychev filters or elliptical filters could be used instead of the Butterworth low-pass filters.

[0119] FIG. 5a is a z-plane pole-zero diagram of an example 4th order falling shelving filter formed according to this method, whilst FIG. 5b is a z-plane pole-zero diagram of an example 4th order rising shelving filter formed according to this method (which is the same as the pole-zero diagram of FIG. 5a, with the poles of FIG. 5a changed into zeros and the zeros of FIG. 5a changed into poles). FIG. 5c is a z-plane pole-zero diagram of an example 5th order falling shelving filter formed according to this method, whilst FIG. 5d is a z-plane pole-zero diagram of an example 5th order rising shelving filter formed according to this method (which is the same as the pole-zero diagram of FIG. 5c, with the poles of FIG. 5c changed into zeros and the zeros of FIG. 5c changed into poles). As is normal, poles are shown as crosses and zeros are shown as circles, and the unit circle 500 is illustrated. The pole-zero diagrams for other integer-order falling shelving filters and rising shelving filters are analogous to those shown in FIGS. 5a-d.

[0120] For a shelving filter generated according to this method, the positions of the zeros and poles for its transfer function are determined according to the positions that would be used for poles for the transfer function of low-pass basis-filters.

[0121] In particular, for an n-th order falling shelving filter with low-corner frequency f₁ and high-corner frequency f₂: (i) the n poles for its transfer function are located at the positions that would be used for poles for an n-th order low-pass basis-filter with a cut-off frequency f₁; and (ii) the n zeros for its transfer function are located at the positions that would be used for poles for an n-th order low-pass basis filter with cut-off frequency f₂.

[0122] On the other hand, for an n-th order rising shelving filter with low-corner frequency f₁ and high-corner frequency f₂: (i) the n poles for its transfer function are located at the positions that would be used for poles for an n-th order low-pass basis-filter with a cut-off frequency f₂; and (ii) the n zeros for its transfer function are located at the positions that would be used for poles for an n-th order low-pass basis filter with cut-off frequency f₁.

[0123] The rate of change of frequency response between the low-corner frequency f₁ and the high-corner frequency f₂ (i.e. the gradient of the frequency response curve between f₁ and f₂ shown in FIGS. 4a and 4b) is determined by the order of the shelving filter and the value of r_B for the basis-type. In general, for an n-th order falling shelving filter, this rate of change of frequency response gain is substantially equal to −r_B×n dB per octave, and for an n-th order rising shelving filter, this rate of change of frequency response gain is substantially equal to r_B×n dB per octave. Thus, when using such an n-th order shelving filter, the gain value g₂ is determined by (i) g₂=g₁−nr_B log₂(f₂/f₁) for a falling shelving filter and (ii) g₂=g₁+nr_B log₂(f₂/f₁) for a rising shelving filter.

[0124] Thus, the gains g₁ and g₂ are interdependent. They can be adjusted together (for example to obtain a desired gain g₁ or a desired gain g₂) by appropriate scaling of the transfer function, as is well-known.

[0125] However, if a shelving filter is required that has an arbitrary low-corner frequency f₁ and an arbitrary high-corner frequency f₂, with arbitrary frequency response gains of g₁ and g₂ at f₁ and f₂ respectively, then it will not always be possible to use a shelving filter generated by Method A as described above, since g₂ is inherently determined by f₁, f₂, g₁, r_B and the order n as described above.

[0126] Embodiments of the invention described below therefore overcome this problem and provide a shelving filter having an arbitrary low-corner frequency f₁ and an arbitrary high-corner frequency f₂, with arbitrary frequency response gains of g₁ and g₂ at f₁ and f₂ respectively. The values of g₁, g₂, f₁ and f₂ may be specified by a user, for example. Using the above equations, the “effective order” n_e for such a shelving filter will thus be defined as

[0127] $n_e=\frac{g_2-g_1}{r_B{\log }_2\left(f_2/f_1\right)},$
where n_e may be a non-integer number. For a shelving filter with effective order n_e, the rate of change of frequency response gain between the low-corner frequency and the high-corner frequency is (i) substantially n_er_B dB per octave for a rising shelving filter or (ii) substantially −n_er_B dB per octave for a falling shelving filter.
(4.2) Method B

[0128] A first method, according to an embodiment of the invention, for generating a shelving filter having an arbitrary low-corner frequency f₁ and an arbitrary high-corner frequency f₂, with arbitrary frequency response gains of g₁ and g₂ at f₁ and f₂ respectively (so that the difference between g₂ and g₁ can be a desired amount), is described below.

[0129] The desired shelving filter is generated by combining two separate shelving filters:

• • [0130] the first of these two shelving filters (i) has a low-corner frequency at f₁ and a high-corner frequency at f₂; and (ii) has an order n that is determined as described below;
  • [0131] the second of these two shelving filters (i) has a low-corner frequency at a frequency f₃ and a high-corner frequency at a frequency f₄, where f₁≦f₃<f₄≦f₂ (i.e. the range from f₃ to f₄ is a sub-range of the range from f₁ to f₂); and (ii) is a 1st order shelving filter;
  • [0132] if g₁>g₂, then the first and second shelving filters are both falling shelving filters; if g₁<g₂, then the first and second shelving filters are both rising shelving filters.

[0133] The first shelving filter may be a shelving filter formed according to method A above, based on a basis-type B. However, it will be appreciated that the first shelving filter may be any type (or class) of shelving filter.

[0134] The order n for the first shelving filter is determined to be the largest integer n for it so that the difference, d, between the gain of the frequency response of the first shelving filter at f₂ and the gain of the frequency response of the first shelving filter at f₁ has a magnitude at most |g₂−g₁|.

[0135] Using the above equations for method A shelving filters of basis-type B, the difference d is d=−nr_B log₂(f₂/f₁) for a falling shelving filter and d=nr_B log₂(f₂/f₁) for a rising shelving filter. Thus, the order for the first shelving filter may be determined according to

[0136] $n=\lfloor \frac{g_2-g_1}{r_B{\log }_2\left(f_2/f_1\right)}\rfloor =\lfloor n_e\rfloor ,$
where, as usual └x┘ represents the highest integer no greater than x.

[0137] The low-corner frequency f₃ and high-corner frequency f₄ for the second shelving filter are determined so that the difference between the frequency response gain of the second shelving filter at f₄ and the frequency response gain of the second shelving filter at f₃ is g₂−g₁−d. It is possible to obtain such values for f₃ and f₄ for the 1st-order second shelving filter because the order n for the first shelving filter was chosen to be the largest integer order for the first shelving filter so that the |d|≦|g₂−g₁|.

[0138] Using the above equations for method A shelving filters of basis-type B, f₃ and f₄ are determined relative to each other such that

[0139] ${\log }_2\left(f_4/f_3\right)=\frac{g_2-g_1-d}{r_B}.$

[0140] The range defined from f₃ to f₄ may be chosen to be anywhere in the range from f₁ to f₂. However, in preferred embodiments, f₃ and f₄ are chosen such that range from f₃ to f₄ is substantially centred in the range from f₁ to f₂. This may be centred in terms of log-frequency, so that f₁f₂=f₃f₄, or in terms of actual frequency, so that f₄+f₃=f₂+f₁. Such centring of the range from f₃ to f₄ within the range from f₁ to f₂ provides improved audio quality as the frequency response resulting from the combination of the first shelving filter and the second shelving filter will retain as sharp a corner as possible at the corner frequencies f₁ and f₂ by such a choice of f₃ and f₄. If instead, for example, f₃ were chosen to be closer to f₁, then the second shelving filter is more likely to have a more significant impact on the frequency response of the resultant shelving filter at frequencies below the low-corner frequency f₁, which is undesirable for the intended resultant shelving filter. Similarly, if f₄ were chosen to be closer to f₂, then the second shelving filter is more likely to have a more significant impact on the frequency response of the resultant shelving filter at frequencies above the high-corner frequency f₂, which is undesirable for the intended resultant shelving filter.

[0141] The combination of the first shelving filter and the second shelving filter is a shelving filter with low-corner frequency f₁ and high-corner frequency f₂, and for which the change in frequency response gain between the frequencies f₁ and f₂ is substantially equal to g₂−g₁.

[0142] The actual gain for the resulting shelving filter is determined by the gains selected for the first and second shelving filters. For example, the gain for the first shelving filter could be set to be g₁ at frequency f₁ and the gain for the second shelving filter could be set to be 0 at the frequency f₃, so that the gain for the combination of the first and the second shelving filters is the desired gain g₁ at frequency f₁. However, it will be appreciated that other gain values could be used for the first and the second shelving filters so that the combination of the these two shelving filters results in the desired gain g₁ at frequency f₁, which then inherently results in the desired gain g₂ at frequency f₂.

[0143] For example, suppose that (i) the basis-type is. Butterworth filters (so that r_B≈6.0); (ii) f₁=100 Hz; (iii) f₂=140 Hz; (iv) g₁=6 dB; and (v) g₂=−3 dB. Then:

• • [0144] the first and second shelving filters will be falling shelving filters (as g₁>g₂);

[0145] $n=\lfloor \frac{g_2-g_1}{r_B{\log }_2\left(f_2/f_1\right)}\rfloor =3,$

• • [0146] meaning that the first shelving filter is a 3rd order filter and the second shelving filter is a 1st order filter;
  • [0147] the frequency response change between f₁ and f₂ resulting from the first shelving filter is d=−nr_B log₂(f₂/f₁)≈−8.7 dB, leaving a change of frequency response of g₂−g₁−d≈−0.3 dB to be attained from the second shelving filter; and hence

[0148] ${\log }_2\left(f_4/f_3\right)=\frac{g_1-g_2-d}{r_B}\approx 0.05$

• • [0149] so that f₄≈1.04f₃. In the preferred embodiment in which f₁f₂=f₃f₄, the values for f₃ and f₄ are about 116 Hz and 121 Hz respectively.

[0150] As another example, suppose that (i) the basis-type is Butterworth filters (so that r_B≈6.0); (ii) f₁=1000 Hz; (iii) f₂=2100 Hz; (iv) g₁=−4 dB; and (v) g₂=7 dB. Then:

• • [0151] the first and second shelving filters will be rising shelving filters (as g₁<g₂);

[0152] $n=\lfloor \frac{g_2-g_1}{r_B{\log }_2\left(f_2/f_1\right)}\rfloor =1,$

• • [0153] meaning that the first shelving filter is a 1st order filter and the second shelving filter is a 1st order filter;
  • [0154] the frequency response change between f₁ and f₂ resulting from the first shelving filter is d=nr_B log₂(f₂/f₁)≈6.4 dB, leaving a change of frequency response of g₂−g₁−d≈4.6 dB to be attained from the second shelving filter; and hence

[0155] ${\log }_2\left(f_4/f_3\right)=\frac{g_1-g_2-d}{r_B}\approx 0.76$

• • [0156] so that f₄≈1.70f₃. In the preferred embodiment in which f₁f₂=f₃f₄, the values for f₃ and f₄ are about 1111 Hz and 1889 Hz respectively.
    (4.3) Method C

[0157] A second method, according to an embodiment of the invention, for generating a shelving filter having an arbitrary low-corner frequency f₁ and an arbitrary high-corner frequency f₂, with arbitrary frequency response gains of g₁ and g₂ at f₁ and f₂ respectively (so that the difference between g₂ and g₁ can be a desired amount), is described below.

[0158] The desired shelving filter is generated by combining two separate shelving filters:

• • [0159] the first of these two shelving filters (i) has a low-corner frequency at f₁ and a high-corner frequency at f₂; and (ii) has an order n that is determined as described below;
  • [0160] the second of these two shelving filters (i) has a low-corner frequency at a frequency f₃ and a high-corner frequency at a frequency f₄, where f₁≦f₃<f₄≦f₂ (i.e. the range from f₃ to f₄ is a sub-range of the range from f₁ to f₂); and (ii) is a 1st order shelving filter;
  • [0161] if g₁>g₂, then the first shelving filter is a falling shelving filter and the second shelving filter is a rising shelving filter; if g₁<g₂, then the first shelving filter is a rising shelving filter and the second shelving filter is a falling shelving filter.

[0162] The first shelving filter may be a shelving filter formed according to method A above, based on a basis-type B. However, it will be appreciated that the first shelving filter may be any type (or class) of shelving filter.

[0163] The order n for the first shelving filter is determined to be the smallest integer n for it so that the difference, d, between the gain of the frequency response of the first shelving filter at f₂ and the gain of the frequency response of the first shelving filter at f₁ has a magnitude at least |g₂−g₁|.

[0164] Using the above equations for method A shelving filters of basis-type B, the difference d is d=−nr_B log₂(f₂/f₁) for a falling shelving filter and d=nr_B log₂(f₂/f₁) for a rising shelving filter. Thus, the order for the first shelving filter may be determined according to

[0165] $n=\lceil \frac{g_2-g_1}{r_B{\log }_2\left(f_2/f_1\right)}\rceil =\lceil n_e\rceil ,$
where, as usual ┌x┐ represents the highest integer no greater than x.

[0166] The low-corner frequency f₃ and high-corner frequency f₄ for the second shelving filter are determined so that the difference between the frequency response gain of the second shelving filter at f₄ and the frequency response gain of the second shelving filter at f₃ is g₂−g₁−d. It is possible to obtain such values for f₃ and f₄ for the 1st-order second shelving filter because the order n for the first shelving filter was chosen to be the smallest integer order for the first shelving filter so that the |d|≧|g₂−g₁|.

[0167] The range defined from f₃ to f₄ may be chosen to be anywhere in the range from f₁ to f₂. However, in preferred embodiments, f₃ and f₄ are chosen such that range from f₃ to f₄ is substantially centred in the range from f₁ to f₂. This may be centred in terms of log-frequency, so that f₁f₂=f₃f₄, or in terms of actual frequency, so that f₄+f₃=f₂+f₁. Such centring of the range from f₃ to f₄ within the range from f₁ to f₂ provides improved audio quality as the frequency response resulting from the combination of the first shelving filter and the second shelving filter will retain as sharp a corner as then possible at the corner frequencies f₁ and f₂ by such a choice of f₃ and f₄. If instead, for example, f₃ were chosen to be closer to f₁, then the second shelving filter is more likely to have a more significant impact on the frequency response of the resultant shelving filter at frequencies below the low-corner frequency f₁, which is undesirable for the intended resultant shelving filter. Similarly, if f₄ were chosen to be closer to f₂, then the second shelving filter is more likely to have a more significant impact on the frequency response of the resultant shelving filter at frequencies above the high-corner frequency f₂, which is undesirable for the intended resultant shelving filter.

[0168] The combination of the first shelving filter and the second shelving filter is a shelving filter with low-corner frequency f₁ and high-corner frequency f₂, and for which the change in frequency response gain between the frequencies f₁ and f₂ is substantially equal to g₂−g₁.

[0169] The actual gain for the resulting shelving filter is determined by the gains selected for the first and second shelving filters. For example, the gain for the first shelving filter could be set to be g₁ at frequency f₁ and the gain for the second shelving filter could be set to be 0 at the frequency f₃, so that the gain for the combination of the first and the second shelving filters is the desired gain g₁ at frequency f₁. However, it will be appreciated that other gain values could be used for the first and the second shelving filters so that the combination of the these two shelving filters results in the desired gain g₁ at frequency f₁, which then inherently results in the desired gain g₂ at frequency f₂.

[0170] For example, suppose that (i) the basis-type is Butterworth filters (so that r_B≈6.0); (ii) f₁=100 Hz; (iii) f₂=140 Hz; (iv) g₁=6 dB; and (v) g₂=−3 dB. Then:

• • [0171] the first shelving filter will be a falling shelving filter and the second shelving filter will be a rising shelving filter (as g₁>g₂);

[0172] $n=\lceil \frac{g_2-g_1}{r_B{\log }_2\left(f_2/f_1\right)}\rceil =4,$
meaning that the first shelving filter is a 4th order filter and the second shelving filter is a 1st order filter;

• • [0173] the frequency response change between f₁ and f₂ resulting from the first shelving filter is d=−nr_B log₂(f₂/f₁)≈−11.7 dB, leaving a change of frequency response of g₂−g₁−d≈2.7 dB to be attained from the second shelving filter; and hence

[0174] ${\log }_2\left(f_4/f_3\right)=\frac{g_1-g_2-d}{r_B}\approx 0.45$
so that f₄≈1.37f₃. In the preferred embodiment in which f₁f₂=f₃f₄, the values for f₃ and f₄ are about 101 Hz and 138 Hz respectively.

[0175] As another example, suppose that (i) the basis-type is Butterworth filters (so that r_B≈6.0); (ii) f₁=1000 Hz; (iii) f₂=2100 Hz; (iv) g₁=−4 dB; and (v) g₂=7 dB. Then:

• • [0176] the first shelving filter will be a rising shelving filter and the second shelving filter will be a falling shelving filter (as g₁<g₂);

[0177] $n=\lceil \frac{g_2-g_1}{r_B{\log }_2\left(f_2/f_1\right)}\rceil =2,$

• • [0178] meaning that the first shelving filter is a 2nd order filter and the second shelving filter is a 1st order filter;
  • [0179] the frequency response change between f₁ and f₂ resulting from the first shelving filter is d=nr_B log₂(f₂/f₁)≈12.8 dB, leaving a change of frequency response of g₂−g₁−d≈−1.8 dB to be attained from the second shelving filter; and hence

[0180] ${\log }_2\left(f_4/f_3\right)=\frac{g_1-g_2-d}{r_B}\approx 0.3$

• • [0181] so that f₄≈1.24f₃. In the preferred embodiment in which f₁f₂=f₃f₄, the values for f₃ and f₄ are about 1307 Hz and 1607 Hz respectively.
    (4.4) Method D

[0182] In alternative embodiments of the invention, the Method B or C is used, except that the order of the second shelving filter may be a different predetermined order m. In this case, the relationship between f₃ and f₄ is determined according to

[0183] ${\log }_2\left(f_4/f_3\right)=\frac{g_2-g_1-d}{{\mathrm{mr}}_B}.$
In this case, the second shelving filter may a shelving filter formed according to method A above, based on the same basis-type B as for the first shelving filter. However, it will be appreciated that the second shelving filter may be any type (or class) of shelving filter.

[0184] Additionally, then, when Method D is based on Method B, the order n for the first shelving filter may be less than that calculated using Method B. For example, the order n for the first shelving filter may be determined to be any integer n so that the difference, d, between the gain of the frequency response of the first shelving filter at f₁ and the gain of the frequency response of the first shelving filter at f₂ has a magnitude at most |G₂−g₁|.

[0185] Additionally, then, when Method D is based on Method C, the order n for the first shelving filter may be greater than that calculated using Method B. For example, the order n for the first shelving filter may be determined to be any integer n so that the difference, d, between the gain of the frequency response of the first shelving filter at f₁ and the gain of the frequency response of the first shelving filter at f₂ has a magnitude at least |g₂−g₁|.

(4.5) Method E

[0186] In alternative embodiments of the invention, the Method D is used, except that the second shelving filter may be based on basis-type, B′, different from the basis-type B for the first shelving filter. The above equations to calculate the relationship between f₃ and f₄ (i.e. to calculate log₂(f₄/f₃)) would use r_B′ as opposed to r_B.

(4.6) Method F

[0187] FIGS. 6a-6i are z-plane pole-zero diagrams for example shelving filters according to a further embodiment of the invention for generating a shelving filter having an arbitrary low-corner frequency f₁ and an arbitrary high-corner frequency f₂, with arbitrary frequency response gains of g₁ and g₂ at f₁ and f₂ respectively (so that the difference between g₂ and g₁ can be a desired amount).

[0188] For a given cut-off frequency f, an order n low-pass basis-filters with that cut-off frequency f has n poles. The set of poles for all such low-pass basis-filters with cut-off frequency f as n ranges from 1 to infinity defines a curve in the z-plane corresponding to that basis-type and that frequency f. In each of FIGS. 6a-6i, the right-hand dashed curve 652 is the curve corresponding to the chosen basis-type and with a cut-off frequency equal to the low-corner frequency f₁, whilst the left-hand dashed curve 650 is the curve corresponding to the chosen basis-type and with a cut-off frequency equal to the high-corner frequency f₂. The unit circle 500 is not shown in FIGS. 6a-6i for clarity purposes.

[0189] The shelving filters illustrated in FIGS. 6a-6i are falling shelving filters. Analogous rising shelving filters can be obtained by interchanging the poles and zeros.

[0190] FIG. 6a is a z-plane pole-zero diagram for a shelving filter with an effective order n_e of 0.

[0191] This shelving filter is formed from a 1st order filter section giving a zero and a pole both located at a position 600 on the real axis 654 halfway between the curve 650 and the curve 652.

[0192] FIG. 6c is a z-plane pole-zero diagram for a shelving filter with an effective order n_e of 1.

[0193] This shelving filter is formed from a 1st order filter section giving a zero and a pole at the positions of the zero and the pole for a 1st order shelving filter formed according to the above-described Method A with low-corner frequency f₁ and high-corner frequency f₂. Thus the zero is located at a position 602 on the real axis 654 at which the curve 650 intersects the real axis 654, and the pole is located at a position 604 on the real axis 654 at which the curve 652 intersects the real axis 654.

[0194] FIG. 6b is a z-plane pole-zero diagram for a shelving filter with an effective order n_e=α, where 0<α<1, so that the rate of change of gain between the low-corner frequency f₁ and the high-corner frequency f₂ is −n_er_B dB per octave.

[0195] This shelving filter is formed from a 1st order filter section giving a zero and a pole located on the real-axis 654. If the distance between the position 600 and the position 602 is D₁ (which also equals the distance between the position 600 and the position 604), then the zero is a distance αD₁ from the position 600 along the real axis 654 towards the position 602, and the pole is a distance αD₁ from the position 600 along the real axis 654 towards the position 604.

[0196] Thus, as α increases from 0 (the configuration in FIG. 6a) to 1 (the configuration in FIG. 6c), the zero moves along the real axis 654 from the position 600 to the position 602 and the pole moves along the real axis 654 from the position 600 to the position 604.

[0197] FIG. 6e is a z-plane pole-zero diagram for a shelving filter with an effective order n_e of 2.

[0198] This shelving filter is formed from:

[0199] (i) A 2nd order filter section giving two complex-valued zeros and two complex-valued poles at the positions of the zeros and the poles for a 2nd order shelving filter formed according to the above-described Method A with low-corner frequency f₁ and high-corner frequency f₂. The zeros are located at positions 606a and 606b on the curve 650 and the poles are located at positions 608a and 608b on the curve 652.

[0200] (ii) A 1st order filter section giving a zero and a pole both located at the position 600 on the real axis 654 halfway between the curve 650 and the curve 652. This 1st order filter section thus has no actual effect on the resulting shelving filter (and so could be omitted). This 1st order filter section is the same as the 1st order filter section in FIG. 6a.

[0201] FIG. 6d is a z-plane pole-zero diagram for a shelving filter with an effective order n_e=1+α, where 0<α<1, so that the rate of change of gain between the low-corner frequency f₁ and the high-corner frequency f₂ is −n_er_B dB per octave.

[0202] This shelving filter is formed from:

[0203] (i) A 2nd order filter section giving two complex-valued zeros and two complex-valued poles. If the distance along the curve 650 between the position 602 and the position 606a is D₂, then one of the complex zeros is positioned a distance αD₂ from the position 602 along the curve 650 towards the position 606a. The other complex-valued zero is positioned a distance αD₂ from the position 602 along the curve 650 towards the position 606b. Similarly, if the distance along the curve 652 between the position 604 and the position 608a is D₃, then one of the complex poles is positioned a distance αD₃ from the position 604 along the curve 652 towards the position 608a. The other complex-valued pole is positioned a distance αD₃ from the position 604 along the curve 652 towards the position 608b.

[0204] (ii) A 1st order filter section giving a real-valued zero and a real-valued pole, with the real-valued zero being a distance αD₁ from the position 604 along the real axis 654 towards the position 600, and the real-valued pole being a distance αD₁ from the position 602 along the real axis 654 towards the position 600.

[0205] The configuration of FIG. 6c can be viewed as having an additional pole and zero at the position 602 (which therefore cancel each other out) and an additional pole and zero at the position 604 (which therefore cancel each other out). Thus, as α increases from 0 (the configuration in FIG. 6c) to 1 (the configuration in FIG. 6e), a real-valued zero moves from the position 604 to the position 600 and a real-valued pole moves from the position 602 to the position 600. Similarly, the complex-valued zeros move along the curve 650 from the positions 602 and 602 to the positions 606a and 606b, and the complex-valued poles move along the curve 652 from the positions 604 and 604 to the positions 608a and 608b.

[0206] FIG. 6g is a z-plane pole-zero diagram for a shelving filter with an effective order n_e of 3.

[0207] This shelving filter is formed from:

[0208] (i) A 2nd order filter section giving two complex-valued zeros and two complex-valued poles at the positions of the complex zeros and the complex poles for a 3rd order shelving filter formed according to the above-described Method A with low-corner frequency f₁ and high-corner frequency f₂. The complex zeros are located at positions 610a and 610b on the curve 650 and the complex poles are located at positions 612a and 612b on the curve 652.

[0209] (ii) A 1st order filter section giving a real-valued zero located at the position 602 on the real axis 654 and a real-valued pole located at the position 604 on the real axis 654. This is the same as the 1st order filter section in FIG. 6c.

[0210] FIG. 6f is a z-plane pole-zero diagram for a shelving filter with an effective order n_e=2+α, where 0<α<1, so that the rate of change of gain between the low-corner frequency f₁ and the high-corner frequency f₂ is −n_er_B dB per octave.

[0211] This shelving filter is formed from:

[0212] (i) A 2nd order filter section having two complex-valued zeros and two complex-valued poles. If the distance along the curve 650 between the position 606a and the position 610a is D₄, then one of the complex zeros is positioned a distance αD₄ from the position 606a along the curve 650 towards the position 610a. The other complex-valued zero is positioned a distance αD₄ from the position 606b along the curve 650 towards the position 610b. Similarly, if the distance along the curve 652 between the position 608a and the position 612a is D₅, then one of the complex poles is positioned a distance αD₅ from the position 608a along the curve 652 towards the position 612a. The other complex-valued pole is positioned a distance αD₅ from the position 608b along the curve 652 towards the position 612b.

[0213] (ii) A 1st order filter section giving a real-valued zero and a real-valued pole, with the real-valued zero being a distance αD₁ from the position 600 along the real axis 654 towards the position 602, and the real-valued pole being a distance αD₁ from the position 600 along the real axis 654 towards the position 604. This is the same as the 1st order filter section for FIG. 6b.

[0214] Thus, as α increases from 0 (the configuration in FIG. 6e) to 1 (the configuration in FIG. 6g), the real-valued zero moves from the position 600 to the position 602 along the real axis 654 and the real-valued pole moves from the position 600 to the position 604 along the real axis 654. Similarly, the complex-valued zeros move along the curve 650 from the positions 606a and 606b to the positions 610a and 610b respectively, and the complex-valued poles move along the curve 652 from the positions 608a and 608b to the positions 612a and 612b respectively.

[0215] FIG. 6i is a z-plane pole-zero diagram for a shelving filter with an effective order n_e of 4.

[0216] This shelving filter is formed from:

[0217] (i) A 2nd order filter section giving two complex-valued zeros and two complex-valued poles at the positions of the two zeros and the two poles closest to the unit circle 500 for a 4th order shelving filter formed according to the above-Method A with low-corner frequency f₁ and high-corner frequency f₂. The zeros are located at positions 613a and 613b on the curve 650 and the poles are located at positions 616a and 616b on the curve 652.

[0218] (ii) A 2nd order filter section giving two complex-valued zeros and two complex-valued poles at the positions of the other two zeros and the other two poles for the 4th order shelving filter formed according to the above-described Method A. The zeros are located at positions 614a and 614b on the curve 650 and the poles are located at positions 618a and 618b on the curve 652.

[0219] (iii) A 1st order filter section giving a zero and a pole both located at the position 600 on the real axis halfway between curve 650 and the curve 652. This 1st order filter section thus has no actual effect on the resulting shelving filter (and so could be omitted). This 1st order filter section is the same as the 1st order filter section in FIGS. 6a and 6e.

[0220] FIG. 6h is a z-plane pole-zero diagram for a shelving filter with an effective order n_e=3+α, where 0<α<1, so that the rate of change of gain between the low-corner frequency f₁ and the high-corner frequency f₂ is −n_er_B dB per octave.

[0221] This shelving filter is formed from:

[0222] (i) A 2nd order filter section giving two complex-valued zeros and two complex-valued poles. If the distance along the curve 650 between the position 610a and the position 613a is D₆, then one of the complex zeros is positioned a distance αD₆ from the position 610a along the curve 650 towards the position 613a. The other complex-valued zero is positioned a distance αD₆ from the position 610b along the curve 650 towards the position 613b. Similarly, if the distance along the curve 652 between the position 612a and the position 616a is D₇, then one of the complex poles is positioned a distance αD₇ from the position 612a along the curve 652 towards the position 616a. The other complex-valued pole is positioned a distance αD₇ from the position 612b along the curve 652 towards the position 616b.

[0223] (i) A 2nd order filter section giving two complex-valued zeros and two complex-valued poles. If the distance along the curve 650 between the position 602 and the position 614a is D₈, then one of the complex zeros is positioned a distance αD₈ from the position 602 along the curve 650 towards the position 614a. The other complex-valued zero is positioned a distance αD₈ from the position 602 along the curve 650 towards the position 614b. Similarly, if the distance along the curve 652 between the position 604 and the position 618a is D₉, then one of the complex poles is positioned, a distance αD₈ from the position 604 along the curve 652 towards the position 618a. The other complex-valued pole is positioned a distance αD₈ from the position 604 along the curve 652 towards the position 618b.

[0224] (ii) A 1st order filter section giving a real-valued zero and a real-valued pole, with the real-valued zero being a distance αD₁ from the position 604 along the real axis 654 towards the position 600, and the real-valued pole being a distance αD₁ from the position 602 along the real axis 654 towards the position 600. This 1st order filter section is the same as the 1st order filter section in FIG. 6d.

[0225] The configuration of FIG. 6g can be viewed as having an additional pole and zero at the position 602 (which therefore cancel each other out) and an additional pole and zero at the position 604 (which therefore cancel each other out). Thus, as α increases from 0 (the configuration in FIG. 6g) to 1 (the configuration in FIG. 6i), a real-valued zero moves from the position 604 to the position 600 along the real axis 654 and a real-valued pole moves from the position 602 to the position 600 along the real axis 654. Similarly, the complex-valued zeros move along the curve 650 from the positions 602, 602, 610a and 610b to the positions 614a, 614b, 613a and 613b respectively, and the complex-valued poles move along the curve 652 from the positions 604, 604, 612a and 612b to the positions 618a, 618b, 616a and 616b respectively.

[0226] The skilled person will appreciate that this method may be extended analogously to higher effective filter orders n_e.

[0227] In each of FIG. 6d onwards:

• • [0228] The 2nd-order filter section(s) together provide a first shelving filter, each 2nd-order filter corresponding to a complex-valued zero (and its complex conjugate) and a complex-valued pole (and its complex conjugate).
  • [0229] The 1st-order filter section that is used provides a second shelving filter, of order 1, the 1st-order filter corresponding to the real-valued zero and the real-valued pole.
  • [0230] The low-corner frequency for the first shelving filter is f₁ and the high-corner frequency for the first shelving filter is f₂.
  • [0231] The order of the first shelving filter is determined as being twice the number of 2nd-order filter sections used. In particular, the order n for the first shelving filter is

[0232] $n=2\lceil \frac{n_e-1}{2}\rceil .$

• • [0233] The positions of the real-valued pole and zero from the 1st-order filter section determine the low-corner frequency f₃ and the high-corner frequency f₄ for the second shelving filter. As the real-values pole and zero are always between the curves 650 and 652, this results in f₁≦f₃≦f₄≦f₂.
  • [0234] When the effective order n_e=w+α, where 0<α<1 and w is an even positive integer, (i) if a falling shelving filter is to be produced, then the first and second shelving filters are both falling shelving filters (ii) if a rising shelving filter is to be produced, then the first and second shelving filters are both rising shelving filters.
  • [0235] When the effective order n_e=w+α, where 0<α<1 and w is an odd positive integer, (i) if a falling shelving filter is to be produced, then the first shelving filter is a falling shelving filter and the second shelving filter is a rising shelving filter (ii) if a rising shelving filter is to be produced, then the first shelving filter is a rising shelving filter and the second shelving filter is a falling shelving filter.

[0236] In general, to generate a Method F shelving filter with effective order n_e=w+α, where 0<α<1 and w is an integer, with low-corner frequency f₁ and high-corner frequency f₂:

[0237] (i) the complex-valued zeros for the shelving filter are at positions interpolated between zeros for a Method A shelving filter of order w with low-corner frequency f₁ and high-corner frequency f₂ and zeros for a Method A shelving filter of order w+1 with low-corner frequency f₁ and high-corner frequency f₂;

[0238] (ii) the complex-valued poles for the shelving filter are at positions interpolated between poles for a Method A shelving filter of order w with low-corner frequency f₁ and high-corner frequency f₂ and the poles for a Method A shelving filter of order w+1 with low-corner frequency f₁ and high-corner frequency f₂; and

[0239] (iii) the real-valued zero and pole are at interpolated positions on the real-axis between the curves 650 and 652. This is interpolation along the real-axis in the log-frequency domain.

[0240] Thus, for a rising shelving filter with low-corner frequency f₁ and high-corner frequency f₂:

[0241] (i) each zero of the transfer function of the first shelving filter is located at a respective location determined by interpolating between (a) the location of a corresponding pole of the transfer function of a first low-pass basis-filter having a cut-off frequency of f₁ and an order of n−1 and (b) the location of a corresponding pole of the transfer function of a second low-pass basis-filter having a cut-off frequency of f₁ and an order of n; and

[0242] (ii) each pole of the transfer function of the first shelving filter is located at a respective location determined by interpolating between (c) the location of a corresponding pole of the transfer function of a third low-pass basis-filter having a cut-off frequency of f₂ and an order of n−1 and (d) the location of a corresponding pole of the transfer function for a fourth low-pass basis-filter having a cut-off frequency of f₂ and an order of n.

[0243] Furthermore, for a falling shelving filter with low-corner frequency f₁ and high-corner frequency f₂:

[0244] (i) each pole of the transfer function of the first shelving filter is located at a respective location determined by interpolating between (a) the location of a corresponding pole of the transfer function of a first low-pass basis-filter having a cut-off frequency of f₁ and an order of n−1 and (b) the location of a corresponding pole of the transfer function of a second low-pass basis-filter having a cut-off frequency of f₁ and an order of n; and

[0245] (ii) each zero of the transfer function of the first shelving filter is located at a respective location determined by interpolating between (c) the location of a corresponding pole of the transfer function of a third low-pass basis-filter having a cut-off frequency of f₂ and an order of n−1 and (d) the location of a corresponding pole of the transfer function for a fourth low-pass basis-filter having a cut-off frequency of f₂ and an order of n.

[0246] The interpolation described above with reference to FIGS. 6a-i for the positions of the complex-valued poles and zeros is interpolation along the curve 650 or 652. However, in another embodiment, this interpolation is performed along a straight line between zeros (or poles respectively) for a Method A shelving filter of order w with low-corner frequency f₁ and high-corner frequency f₂ and zeros (or poles respectively) for a Method A shelving filter of order w+1 with low-corner frequency f₁ and high-corner frequency f₂.

[0247] In a preferred embodiment, when multiple 2nd-order filter sections are used (for example in FIGS. 6h and 6i), then a first 2nd-order filter section is based on the two poles and the two zeros that are nearest to the unit circle 500, a second 2nd-order filter section is based on the two poles and the two zeros next nearest to the unit circle 500, and so on. Thus, if s 2nd-order filter sections are used (for s>1), then the t-th 2nd order filter section (1≦t≦s) is based on the pole (and its complex conjugate) that is the t-th closest complex pole to the unit circle 500 and on the zero (and its complex conjugate) that is the t-th closest complex zero to the unit circle 500.

[0248] This is schematically illustrated in FIG. 9, which schematically illustrates zeros z₁ . . . z_s and their complex conjugate zeros z₁ . . . z_s and poles p₁ . . . p_s and their complex conjugate poles p₁ . . . p for a shelving filter, in which zero z₁ is closer to the unit circle 500 than zero z_j if i>j, and in which pole p_i is closer to the unit circle 500 than pole p_j if i>j. The 2nd-order filter sections in this preferred embodiment are essentially individual filters that are formed by the pairings (z_i, z_i, p_i, p_i), with respective individual transfer functions of

[0249] $\frac{\left(z-z_i\right)\left(z-{\stackrel{\_}{z}}_i\right)}{\left(z-p_i\right)\left(z-{\stackrel{\_}{p}}_i\right)}.$

[0250] Such a formation of 2nd-order filter sections assists in the transition from one set of generated filters to another set of generated filters at the step S206 after the step S208 has been used. This will be described in more detail later.

[0251] However, it will be appreciated that the 2nd-order filter sections need not be generated in this particular manner.

(5) Determining a Set of Filters

[0252] FIG. 7 is a flowchart schematically illustrating the processing 700 according to an embodiment of the invention for determining a set of filters at the step S204. The plurality of filters is determined based upon the currently specified set of frequency response indicators, which may be either the set of frequency response indicators as initialised at the step S202 or may be the set of frequency response indicators as updated at the step S208.

[0253] The plurality of frequency response indicators to use (i.e. the plurality of frequency response indicators initialised at the step S202 or updated at the step S208 as appropriate) shall be referred to below as the set {(F_n,G_n): n=1 . . . N}, where: (i) there are N frequency response indicators, with N≧0; (ii) for the n-th frequency response indicator (where the corresponding user-defined frequency is F_n and the user-defined gain is G_n; and (iii) for convenience of notation, it is assumed that F_n<F_n+1 for 1≦n<N. Two frequency response indicators (F_n,G_n) and (F_m,G_m) shall be referred to as adjacent frequency response indicators if m=n+1 or n=m+1 in the above ordering, i.e. there is no frequency response indicator that specifies a frequency between F_n and F_m.

[0254] At a step S702, it is determined whether N equals 0, i.e. whether the set of frequency response indicators is empty. If N equals 0, then processing continues at a step S704, at which the set of filters to use is set to be empty (so that at the step S206, the audio data would be unaffected by the processor 114) and then processing continues to a step S724, to be described later. Otherwise, if N is not equal to 0, so that there is at least one frequency response indicator, then processing continues at a step S706.

[0255] At the step S706, it is determined whether N equals 1, i.e. whether a single frequency response indicator has been specified. If N equals 1, then processing continues at a step S708, at which a filter with a frequency response having a constant gain equal to G₁ is generated (methods for doing this being well-known in this field of technology) and then processing continues to the step S724, to be described later. Alternatively, the processing at the step S708 may be the same as at the step S704 (i.e. no filter is generated), in which case filters will only be generated once at least two frequency response indicators have been specified.

[0256] If N is not equal to 1, so that there is there is a plurality of frequency response indicators, then processing continues at a step S710. At the step S710, a value i is initialised to the value 1. The processing 700 may make use of one or more iterations in order to determine the set of filters and the value i is an iteration-counter representing the number of the current iteration.

[0257] Next, at a step S712, for each frequency response indicator (F_k,G_k), a corresponding target gain T_k is initialised to be the user-specified gain G_k. The purpose of the target gains T_k will be described later.

[0258] Next, at an optional step S714, for each pair of adjacent frequency response indicators (F_k,G_k) and (F_k+1,G_k+1), a corresponding frequency-proportion h_k is initialised to a predetermined initialisation value. The purpose of the frequency-proportions h_k will be described later. A value of ⅓ has been found to be a suitable predetermined initialisation value for the frequency-proportions h_k, as discussed in more detail later.

[0259] At an optional step S716, the frequency proportions h_k are adjusted. This will be described in more detail later. However, in some embodiments, the step S716 is omitted.

[0260] At an optional step S718, the target gains T_k are adjusted. This will be described in more detail later. However, in some embodiments, the step S718 is omitted.

[0261] At a step S720, a shelving filter is determined for each adjacent pair of frequency response indicators (F_k,G_k) and (F_k+1,G_k+1). Some embodiments of the invention generate a shelving filter for an adjacent pair of frequency response indicators (F_k,G_k) and (F_k+1,G_k+1) using one of the Methods A-F described above.

[0262] For an adjacent pair of frequency response indicators (F_k,G_k) and (F_k+1,G_k+1), the corresponding shelving filter is determined as follows:

• • [0263] In an embodiment that does not make use of the frequency-proportions h_k, the low-corner frequency and the high corner frequency to use for generating the corresponding shelving filter are F_k and F_k+1 respectively. In particular, if one of the above-described Methods A-F is used, then f₁=F_k and f₂=F_k+1.
  • [0264] In an embodiment that does make use of the frequency-proportions h_k, the low-corner frequency f₁ and the high corner frequency f₂ to use for generating the corresponding shelving filter are the values F_k and F_k+1 respectively, adjusted according to the value h_k. The Methods A-F may then be used to generate the shelving filter using the low-corner frequency f₁ and the high corner frequency f₂.
    • [0265] In particular, if h_k is less than 1, then the range from f₁ to f₂ lies within the range from F_k to F_k+1; whereas if h_k is greater than 1, then the range from F_k to F_k+1 lies within the range from f₁ to f₂; and if h_k=1 then f₁=F_k and f₂=F_k+1.
    • [0266] In one embodiment of the invention, f₁ and f₂ are chosen such that the size of the range from f₁ to f₂ is h_k times the size of the range from F_k to F_k+1, so that f₂−f₁=h_k(F_k+1−F_k). Preferably then, if h_k is less than 1, then the range from f₁ to f₂ is substantially centred in the range from F_k to F_k+1; whereas if h_k is greater than 1, then the range from F_k to F_k+1 is substantially centred in the range from f₁ to f₂. Thus, then, preferably f₂+f₁=F_k+1+F_k. From these two equations, the values of f₁ to f₂ can be determined.
    • [0267] In another embodiment of the invention, f₁ and f₂ are chosen such that the size of the range from log f₁ to log f₂ is h_k times the size of the range from log F_k to log F_k+1, so that
    • [0268] log f₂−log f₁=h_k(log F_k+1−log F_k), i.e. f₂/f₁=(F_k+1/F_k)^hk. Preferably then, if h_k is less than 1, then the range from log f₁ to log f₂ is substantially centred in the range from log F_k to log F_k+1; whereas if h_k is greater than 1, then the range from log F_k to log F_k+1 is substantially centred in the range from log f₁ to log f₂. Thus, then, preferably log f₂+log f₁=log F_k+1+log F_k, i.e. f₂f₁=F_k+1F_k. From these two equations, the values of f₁ to f₂ can be determined.
    • [0269] The purpose of using the frequency-proportions will be described later.
  • [0270] The shelving filter is generated using the target gain T_k at the above-determined low-corner frequency f₁ and the target gain T_k+1 at the above-determined high-corner frequency f₂.
  • [0271] If k=1, then the shelving filter is generated with the aim of having a gain of T₁ at frequency F₁ and a change of gain of T₂−T₁ as the frequency changes from F₁ to F₂. If k>1, then the shelving filter is generated with the aim of having a gain of 0 at frequency F_k and a change of gain of T_k+1−T_k as the frequency changes from F_k to F_k+1. Thus, the actual gains achieved at the frequency response indicators will be substantially close to the target values T_k. However, it will be appreciated that the individual shelving filters could be generated with different gains, but still have the same effective combined gains.

[0272] Once the shelving filters have been generated at the step S720, then at a step S722, it is determined whether the iteration-counter i is now a predetermined maximum number of iterations i_Max. If the iteration-counter i equals the predetermined maximum number of iterations i_Max, then processing terminates at the step S724; otherwise, processing continues at a step S726. In this way, the number of iterations performed for the processing 700 for determining the set of filters for the step S204 is at most this predetermined maximum number of iterations. The predetermined maximum number of iterations may be chosen to be a value such that the processing 700 of FIG. 7 is usually expected to have attained, after that predetermined maximum number of iterations, a set of filters whose frequency response is sufficiently close to the desired frequency response, i.e. the amount by which the frequency response resulting from the generated set of filters differs from the desired frequency response does not normally change significantly after i_Max iterations have been performed. A value for i_Max in the range from 3 to 6 has been found to be normally sufficient for this criterion, although values up to 10 or even higher could be used. Alternatively, or additionally, the predetermined maximum number of iterations may be chosen based on the processing capabilities of the processor 114 and/or the amount of processing time to be allowed for performing the iterations. In particular, the value of i_Max may be reduced if the processing capabilities of the processor 114; or the allowed processing time, are relatively low, whereas the value of i_Max may be increased if the processing capabilities of the processor 114, or the allowed processing time, are relatively high. In this way, real-time processing of audio data at the step S206 can be performed with real-time generation of the set of filters, since the maximum number of iterations for the processing 700 can be set to be a suitable value so as not to cause too significant a delay in the generation of the set of filters to use to process the audio data.

[0273] If the current iteration is the last iteration, then processing will have progressed from the step S722 to the step S724. At the step S724, the coefficients for the generated filters may be calculated (methods for doing this being well-known in this field of technology). Additionally, if a visual display of the frequency response of the combination of the set of filters is to be displayed (for example, the curve 304 on the GUI 301), then that frequency response may also be calculated and then displayed (methods for doing this being well-known in this field of technology).

[0274] If the current iteration is not the last iteration, then processing will have progressed from the step S722 to the step S726. At the step S726, for each frequency F_k associated with a frequency response indicator, the processor 114 determines the corresponding frequency response at that frequency F_k resulting from the combination of the determined set of filters. Methods for doing this are well known and shall therefore not be described in detail herein. In the following description, the gain of the frequency response resulting from the combination of the set of filters at frequency F_k will be denoted by A_k.

[0275] Then, at a step S728, an error value e_k for each of the frequency response indicators is determined, where e_k=A_k−G_k for k=1 . . . N.

[0276] At an optional step S730, it is determined whether the error (or difference) between the desired frequency response and the frequency response resulting from the combination of the determined set of filters is sufficiently small. This may be achieved in a variety of ways. For example, one or more of the following methods may be used:

• • [0277] The error (or difference) between the desired frequency response and the frequency response resulting from the combination of the determined set of filters is considered to be sufficiently small if the magnitude of each error value e_k is less than a predetermined error threshold E, and is not sufficiently small otherwise.
  • [0278] The error (or difference) between the desired frequency response and the frequency response resulting from the combination of the determined set of filters is considered to be sufficiently small if

[0279] ${\left(\frac{1}{N}\sum _{k=1}^N{e}_k^p\right)}^{\frac{1}{p}}<E$
(for a positive value of p, such as p=2, and a predetermined error threshold E), and is not sufficiently small otherwise. This example above involves calculating a particular average of the error values e_k. For example, when p=1, then the arithmetic mean of the error values e_k is used, whilst if p=2, then the RMS of the error values e_k is used. However, it will be appreciated that other averages (such as mode, median, etc.) could be used.

[0280] It will be appreciated that embodiments of the invention may use any other measure of the difference between the frequency response resulting from the combination of the generated set of filters and the desired frequency response.

[0281] If it is determined that the error (or difference) between the desired frequency response and the frequency response resulting from the combination of the determined set of filters is sufficiently small, then processing terminates at the step S724; otherwise, processing continues to a step S732. Thus, for embodiments that use the optional step S730, the processing 700 may terminate without having to perform the maximum number i_Max of iterations, thereby saving on processing time, when the combination of the determined set of filters has a frequency response sufficiently close to the desired frequency response (i.e. when further iterations would be deemed unnecessary).

[0282] A relatively high predetermined error threshold E may be used in the above examples in order to reduce the number of iterations that may be required to generate the set of filters, but in doing so potentially producing a frequency response that is more different from the desired frequency response than may otherwise have been attainable with additional iterations. In contrast, a relatively lower predetermined error threshold E may be used in order to try to attain a frequency response that is closer to the desired frequency response, but potentially at the expense of increasing the number of iterations that are then required to generate the set of filters. A predetermined error threshold E of around 0.1 dB may be used for these purposes, as differences in frequency response gain below this level are usually not discernable by a listener.

[0283] At the step S732, the iteration-counter i is incremented to reflect the fact that another iteration is about to be performed.

[0284] Then, at a step S734, one or more target gains T_k is updated. The processor 114 may adjust one or more of the target gains T_k so that the frequency response gain of the combination of the set of filters generated at the step S720 during the next iteration of the processing 700 at the corresponding frequency F_k is closer to the desired gain G_k than for the currently generated set of filters. In other words, the processor attempts to adjust one or more of the target gains T_k so as to try to produce a smaller value of |e_k| during the next iteration of the processing 700.

[0285] In one embodiment, each target gain T_k (for k=1 . . . N) is adjusted so that it assumes the value T_k-αe_k for a positive value α. The value of α may be a predetermined value. However, in other embodiments of the invention, the value of α may vary depending on the current iteration number, for example, the value of α for the i-th iteration could be proportional, or inversely-proportional, to the value i. In other embodiments of the invention, the value of α may vary depending on the value of one or more of the values |e_k|, for example, the value of α for target gain T_k could be a function of the value |e_k|.

[0286] In preferred embodiments, though, the value of α is the constant 1. Thus, for example, if the k-th user-specified gain G_k is 6.0 dB and the gain achieved at the corresponding frequency F_k by the combination of the set of generated filters is A_k=7.2 dB, then the k-th error value is e_k=1.2 dB. Thus, at the end of the first iteration, the target gain value T_k may be adjusted from 6.0B down to 4.8 dB, in the attempt to reduce A_k down to 6.0 dB (i.e. the user-specified gain G_k) in the second iteration. If, at the end of the second iteration, A_k=5.7 dB, then the target gain value T_k may be adjusted from 4.8 dB up to 5.1 dB. This process continues in this fashion for each of the target gains T_k.

[0287] After the target gains T_k have been updated at the step S734, processing returns to the step S714 so that the next iteration for the processing 700 can be carried out.

[0288] FIGS. 10a-g are example graphs 301 of the frequency response curve 304 for the set filters generated at seven successive iterations during the processing 700 for a given set of frequency response indicators 302. FIG. 10a is the frequency response after the 1st iteration; FIG. 10b is the frequency response after the 2nd iteration; FIG. 10c is the frequency response after the 3rd iteration; FIG. 10d is the frequency response after the 4th iteration; FIG. 10e is the frequency response after the 5th iteration; FIG. 10f is the frequency response after the 6th iteration; and FIG. 10g is the frequency response after the 7th iteration.

[0289] As can be seen, after the 1st iteration, the frequency response curve 304 misses most of the frequency response indicators 302. However, from the 3rd iteration onwards, the frequency response curve 304 passes through all of the frequency response indicators 302 virtually exactly.

[0290] It will be appreciated that some embodiments of the invention may impose an upper bound on the order of each shelving filter generated at the step S716. This may be due to limitations on the amount of hardware available for implementing such a shelving filter (for example in a DSP), or may be due to wanting to limit the delay involved in audio processing, since a larger order filter inherently results in a longer processing delay, or may be due to limitations on the amount of processing (and therefore power) that is required (the larger the order of a filter, the greater the amount of processing required to apply that filter). Alternatively, or additionally, and for the same reasons, some embodiments of the invention may impose an upper bound on the sum of all of the orders of the shelving filters that are generated at the step S720. As the orders of the filters generated at the step S720 are dependent on the target gains T_k and the low- and high-corner frequencies used for the shelving, filters, the target gains and the low- and high-corner frequencies may be adjusted in an attempt to ensure that the shelving filters that are generated have appropriately sized orders given the above-mentioned constraint(s). This is the purpose of the optional steps S716 and S718, which may be applied in isolation of, in or combination with, each other. Naturally, the step S716 cannot be applied if the frequency-proportions h_k are not being used.

[0291] FIG. 8a is a schematic flow-chart illustrating processing 800 for the step S718 when an upper bound O_max is imposed on the order of each shelving filter generated at the step S720.

[0292] At a step S802, a counter n is initialised to a value of 1.

[0293] At a step S803, a set of adjustments values d_k (k=1 . . . N) is initialised to zero (i.e. d_k=0 for k=1 . . . N). Each target gain T_k has a corresponding adjustment value d_k that will be calculated during the processing 800 and applied to the target gain T_k at the end of the processing 800.

[0294] At a step S804, it is determined whether n is the value N (i.e. the number of frequency response indicators). If so, then each shelving filter to be generated at the step S720 will have been considered by the processing 800 and so processing terminates at the step S806, at which the target gains T_k are adjusted by their corresponding adjustment values d_k, i.e. T_k=T_k+d_k for k=1 . . . N. Otherwise, processing continues at a step S808.

[0295] At the step S808, the order O_n of the shelving filter that would be generated at the step S720 for the pair of adjacent frequency response indicators (F_n,G_n) and (F_n+1,G_n+1) is determined. The order O_n may be determined, for example, using the above-described equations with reference to Methods A-F as appropriate. This determination is based on using the current corresponding target gains T_n and T_n+1 for these adjacent frequency response indicators, i.e. the target gains T_n and T_n+1 have not been updated yet during this processing 800.

[0296] At a step S810, the required order O_n is compared with the upper bound O_max. If the required order O_n is not greater than the upper bound O_max, then no adjustment of the target gains T_n and T_n+1 is required based on this shelving filter, and so processing continues at a step S812; otherwise processing continues at a step S814.

[0297] At the step S812, the counter n is incremented by 1, and then processing returns to the step S804.

[0298] Alternatively, the step S814 is reached when the required order O_n would be greater than the imposed upper bound O_max on the order for the shelving filters. As such, the target gains T_n and T_n+1 should be adjusted (at the end of the processing 800 at the step S806) accordingly. Thus, at the step S814, the amount R by which the target gains T_n and T_n+1 should be brought together (so that that the order of the required shelving filter is reduced) is determined. Here, R=|T_n−T_n+1|−Ĝ, where Ĝ is the magnitude of the change in gain achievable if the order of this shelving filter were set to be O_max. Bringing the target gains T_n and T_n+1 closer together reduces the magnitude of the gradient of the frequency response curve between the low- and high-corner frequencies, and hence reduces the order of the shelving filter.

[0299] Then, at a step S816, it is determined whether T_n is greater than T_n+1. If T_n is greater than T_n+1, then processing continues at a step S818, at which d_n is decreased by R/2 and d_n+1 is increased by R/2. If T_n is, not greater than T_n+1, then processing continues at a step S820, at which d_n is increased by R/2 and d_n+1 is decreased by R/2. However, it will be appreciated that d_n and d_n+1 may be adjusted by other amounts to attempt to bring T_n and T_n+1 sufficiently close together. For example: at the step S818, d_n may be decreased by vR and d_n+1 may be increased by (1−v)R, and at the step S818, d_n may be increased by vR and d_n+1 may be decreased by (1−v)R, where 0≦v≦1. Processing then continues at the step S812.

[0300] FIG. 8b is a schematic flow-chart illustrating processing 850 for the step S716 when an upper bound O_max is imposed on the order of each shelving filter generated at the step S720. The processing 850 is similar to the processing 800 and steps that they share in common are given the same reference numeral and will therefore not be described again.

[0301] The processing 850 of FIG. 8b begins at the step S802 of FIG. 8a and proceeds to the step S804 of FIG. 8a (without using the step S803). If, at the step S804, n=N, then processing terminates at a step S852; otherwise, processing continues at the step S808 of FIG. 8a and then on to the step S810 of FIG. 8a.

[0302] At the step S810, the required order O_n is compared with the upper bound O_max. If the required order O_n is not greater than the upper bound O_max, then no adjustment of the frequency-proportion h_n is required for the current shelving filter, and so processing continues at the step S812; otherwise processing continues at a step S854.

[0303] At the step S854, the processor 114 increases the value of h_n. This effectively widens the frequency gap between the low-corner frequency and the high-corner frequency for the shelving filter for this pair of frequency response indicators (F_n,G_n) and (F_n+1,G_n+1). This in turn may reduce the order of that shelving filter. Some embodiments of the invention may impose a predetermined upper bound on the frequency-proportion h_n, so that the step S854 cannot increase h_n beyond that upper bound. In some embodiments, this upper bound may be approximately the value 1.

[0304] The step S854 may reduce the order of the shelving filter for the pair of frequency response indicators (F_n,G_n) and (F_n+1,G_n+1) sufficiently to bring it down to the predetermined maximum order O_max. However, for embodiments that impose an upper bound on the frequency-proportion h_n, the step S854 may not reduce the order of the shelving filter sufficiently. Hence, the step S718 of the processing 700 in FIG. 7 may then be used to further reduce the order of the shelving filter.

[0305] As mentioned above, the frequency-proportions h_k are initialised to a predetermined value at the step S714. If a frequency-proportion h_k for a pair of frequency response indicators (F_k,G_k) and (F_k+1,G_k+1) is small (i.e. much less than 1), then the shelving filter produced will have a correspondingly high order but with very little “corner-effect” (i.e. wobbles and curves in the frequency response) at and below the frequency F_k and at and above the frequency F_k+1, since the corner frequencies f₁ and f₂ are moved significantly away from the frequencies F_k and F_k+1. The frequency response of the shelving filter is thus very likely to have a gain of G_k at and below frequency F_k and a gain of G_k+1 at and above frequency F_k+1.

[0306] On the other hand, if a frequency-proportion h_k for a pair of frequency response indicators (F_k,G_k) and (F_k+1,G_k+1) is large (e.g. greater than 1), then the shelving filter produced will have a correspondingly lower order but with increased “corner-effect” (i.e. wobbles and curves in the frequency response) at and below the frequency F_k and at and above the frequency F_k+1, since the corner frequencies f₁ and f₂ are moved towards, and then beyond, the frequencies F_k and F_k+1. The frequency response of the shelving filter is thus less likely to have a gain of G_k at and below frequency F_k and a gain of G_k+1 at and above frequency F_k+1.

[0307] Thus, the choice of the predetermined value at which to initialise the frequency-proportions h_k is a trade-off between the order of the shelving filter (and hence the amount of power, processing, hardware and delay involved) versus the degree of “corner-effect” around the frequencies F_k and F_k+1 and the amount by which the frequency response of the shelving filter deviates from the desired gain of G_k at and below frequency F_k and the desired gain of G_k+1 at and above frequency F_k+1.

[0308] A value of h_k in the range from about ⅓ to about ½ has been found to provide a good compromise, which provides a frequency response for the shelving filter with a relatively constant change of gain per octave between the frequency F_k and the frequency F_k+1 (i.e. a straight line on the log-frequency vs gain frequency response plots), and with a relatively small amount of “corner effect” around the frequency F_k and the frequency F_k+1. Thus, when several frequency response indicators are specified in a straight line on the log-frequency vs gain graph 301, the use of frequency-proportions in this range generate a frequency response curve 304 that is also substantially a straight line through these frequency response indicators.

[0309] This is illustrated in FIGS. 11a-i, which are example graphs 301 of the frequency response curve 304 for a set filters generated during the processing 700 for a given set of frequency response indicators 302a-e. For FIGS. 11a-i, the values of h_k are fixed (i.e. they are not adjusted at the step S716 of FIG. 7) and the target gains T_k are adjusted at the step S718 of FIG. 7. In FIG. 11a, h_k is fixed at 1/12; in FIG. 11b, h_k is fixed at ⅙; in FIG. 11c, h_k is fixed at ¼; in FIG. 11d, h_k is fixed at ⅓; in FIG. 11e, h_k is fixed at ½; in FIG. 11f, h_k is fixed at ⅔; in FIG. 11g, h_k is fixed at 1; in FIG. 11h, h_k is fixed at 3/2; and in FIG. 11i, h_k is fixed at 2.

[0310] The frequency response indicators 302b and 302c have the same associated gain, whilst the frequency response indicators 302c, 302d and 302e form a straight line on the graph 301.

[0311] As can be seen:

• • [0312] for values of h_k of 1/12, ⅙ and ¼ in FIGS. 11a-c:
    • [0313] the frequency response curve 304 between the lined-up frequency response indicators 302c, 302d and 302e is a stepped curve, rather than being the desired substantially straight line; and
    • [0314] the frequency response curve 304 between the level frequency response indicators 302b and 302c is relatively flat.
  • [0315] for values of h_k of ⅔, 1, 3/2 and 2 in FIGS. 11f-i:
    • [0316] the frequency response curve 304 between the lined-up frequency response indicators 302c, 302d and 302e is a relatively straight line, as opposed to the stepped curve of FIGS. 11a-c;
    • [0317] the frequency response curve 304 between the level frequency response indicators 302b and 302c is quite exaggerated;
    • [0318] the frequency response at frequencies below the frequency of the left-most frequency response indicator 302a has a gain substantially different from the gain associated with that frequency response indicator 302a; and
    • [0319] the frequency response at frequencies above the frequency of the right-most frequency response indicator 302e has a gain substantially different from the gain associated with that frequency response indicator 302e.
  • [0320] for values of h_k of ⅓ and ½ in FIGS. 11d-e:
    • [0321] the frequency response curve 304 between the lined-up frequency response indicators 302c, 302d and 302e is a relatively straight line, as opposed to the stepped curve of FIGS. 11a-c;
    • [0322] the frequency response curve 304 between the level frequency response indicators 302b and 302c is relatively flat;
    • [0323] the frequency response at frequencies below the frequency of the left-most frequency response indicator 302a has a gain substantially equal to the gain associated with that frequency response indicator 302a; and
    • [0324] the frequency response at frequencies above the frequency of the right-most frequency response indicator 302e has a gain substantially equal to the gain associated with that frequency response indicator 302e.

[0325] For these reasons, the preferred value for initialising each frequency proportion h_k at the step S714 lies in the range from about ⅓ to about ½.

[0326] The frequency-proportions h_k, the predetermined value at which to initialise the frequency-proportions h_k, and the upper-bound for h_k (if used) may be set by a user through the GUI 300.

(6) Transitioning Between Filters

[0327] The set of filters that have been generated may be implemented in many ways, as is well-known in this field of technology. This implementation may involve implementing one or more of the filters in software or in hardware. For example, each shelving filter could be implemented by determining the coefficients for the linear difference equation for that filter (based on its zeros and poles and the desired gains) and then implementing that linear difference equation. Additionally, the whole set of shelving filters could be implemented by determining the coefficients for the linear difference equation for the combination of the shelving filters (based on all the zeros and poles and desired gains for all of the shelving filters) and then implementing that linear difference equation. However, preferred embodiments of the invention implement each shelving filter as a chain of 2nd-order filter sections (possibly with a single 1st-order filter section if the order of the shelving filter is odd), where each 2nd-order filter section is, in effect, a separate filter whose transfer function has, as its poles and zeros, a corresponding pair of complex-zeros and a corresponding pair of complex-poles from the transfer function of that shelving filter.

[0328] As mentioned above with reference to FIG. 2, once the set of frequency response indicators has been updated at the step S208 and a new set of shelving filters has been generated at the step S204, the processor 114 changes from processing audio data using the previous set of filters to processing the audio data using the updated set of filters.

[0329] In embodiments that implement the shelving filters using the corresponding linear difference equation, the transition from a previous shelving filter to a current shelving filter may be achieved over a period of time by interpolating from the coefficients of the linear difference equation for the previous shelving filter to the coefficients of the linear difference equation for the current shelving filter. Similarly, this may be performed in embodiments that implement the combination of the set of shelving filters by using the corresponding linear difference equation.

[0330] However, the preferred embodiments that implement a shelving filter using a chain of 2nd-order filter sections (possibly with a single 1st-order filter section if the order of the shelving filter is odd) may use a number of measures (described below) to help reduce the likelihood of introducing adverse audio noise as a result of this transition (i.e. to help maintain a high quality level for the processed audio). These measures described below. These measures are particularly suited to the shelving filters generated according to the above-described Method F.

[0331] We note first that the 2nd-order filter sections can be ordered in terms of distance of the zeros and poles from the unit circle 500. In this ordering the i-th 2nd-order filter section is formed from the zero (and its complex conjugate) that is the i-th closest zero to the unit circle 500 and the pole (and its complex conjugate) that is the i-th closest pole to the unit circle 500. Thus, referring to the example shown in FIG. 9, this ordering of the 2nd-order filter sections means that: (a) the 1st 2nd-order filter section is formed from the tuple (z_s, z_s, p_s, p_s); (b) the 2nd 2nd-order filter section is formed from the tuple (z_s−1, z_s−1, p_s−1, p_s−1); and (c) in general, the i-th 2nd-order filter section is formed from the tuple (z_s−1, z_s−i+1, p_s−i+1, p_s−i+1)

[0332] Embodiments of the invention may therefore use one or more of the following measures to help maintain a high quality level for the processed audio when the processing of the audio data transitions from using a previous shelving filter to an updated shelving filter:

• • [0333] (a) When the number of 2nd-order filter sections is the same for the previous shelving filter as the new shelving filter, then the i-th 2nd-order filter section for the new shelving filter inherits, or takes on, the current state-variables for the i-th 2nd-order filter section for the previous shelving filter. The state-variables for a 2nd-order filter section are the input values passed to the 2nd-order filter section and, as the 2nd-order filter section is an infinite impulse response filter, also the output values from the 2nd-order filter section. Thus, if the linear difference equation for the 2nd-order filter section is y[n]=b₀x[n]+b₁x[n−1]+b₂x[n−2]−a₁y[n−2] for input samples x[n] and output samples y[n] and filter coefficients b_i and a_i, then the state variables are the values x[n], x[n−1], x[n−2], y[n−1] and y[n−2]. This helps ensure that the most appropriate state-variables are passed between previous and new 2nd-order filter sections for the shelving filter.
  • [0334] (b) The 2nd-order filter sections for the new shelving filter are processed in the same order (based on the mapping in item (a) above) as the 2nd-order filter sections for the previous shelving filter.
  • [0335] (c) When the number (s) of 2nd-order filter sections for the previous shelving filter is different from the number (t) of 2nd-order filter sections for the new shelving filter, (an example of such a new whelving filter is shown in FIG. 13 in an analogous way to FIG. 9), then the i-th 2nd-order filter section for the new shelving filter inherits the current state-variables for the i-th 2nd-order filter section for the previous shelving filter for i=1 . . . min(s,t). This helps ensure that the most appropriate state-variables are passed between previous and new 2nd-order filter sections. This also helps ensure that any added 2nd-order filter sections (when t>s) and any removed 2nd-order filter sections (when s>t) are formed from zeros and poles that are the furthest from the unit circle 500 and hence have the least audible effect and ability to generate unwanted audio effects/noise.
  • [0336] (d) In item (c) above, when t>s (i.e. when at least one new 2nd-order filter section is added), the state-variables at the new 2nd-order filter sections are initialised to 0 and the coefficients for the new 2nd-order filter sections are set to the required values (based on the poles and zeros associated with the new 2nd-order filter sections). The new shelving filter is then allowed to run for a period of time before the new 2nd-order filter sections are introduced and used, i.e. the new 2nd-order filter sections are initialised to have state-variables of 0, but then receive the audio samples for this period of time in order to establish their state-variables before being introduced (by crossfading) and actually used in the chain of 2nd-order filter sections for the shelving filter. This period of time may be around 0.01 to 0.2 seconds. Preferably, this period of time is 1024 samples for audio data sampled at 44.1 kHz.

[0337] Where the coefficients for the i-th 2nd-order filter section of the previous shelving filter are different from the coefficients for the i-th 2nd-order filter section of the current shelving filter, then the processor 114 updates the coefficients for this 2nd-order filter section over a period of time (such as around 0.01 seconds to 0.2 seconds) by interpolating the coefficients from the coefficients for the previous 2nd-order section to the coefficients for the new 2nd-order section. This, of course, does not apply to newly generated 2nd-order sections.

[0338] It will be appreciated that the above-mentioned techniques (a)-(e) may be applied to other types of filter too, and not just shelving filters, that are implemented using 2nd-order filter sections.

(7) Global Intensity Control

[0339] The GUI 300 may be provided with a global intensity control 322. With this control 322, the user may scale all of the gains G_k for the frequency response indicators (F_k,G_k) by a variable scale value β. The scale value β may vary in a range from 0 to a predetermined maximum value β_max. Using this control 322, all of the gains G_k are changed to become a gain value of βG_k, so that the gains G_k are multiplied by this scale factor β. Setting β=1 sets all of the gains G_k to be original user-specified gain. Setting β<1 reduces the overall intensity of the filtering, and setting β=0 sets all of the gains G_k to be zero, so that no equalisation or filtering is performed at all. Setting β>1 increases the overall intensity of the filtering.

[0340] In this way, the user can easily control the amount of equalisation being applied, without having to adjust each of the frequency response indicators 302 individually.

[0341] It will be appreciated that, insofar as embodiments of the invention are implemented by a computer program, then a storage medium and a transmission medium carrying the computer program form aspects of the invention.