Method and apparatus for percussive noise reduction in a conference

[0001] The subject matter of the present disclosure relates to a method and apparatus for reducing percussive noise that may occur during a conference.

[0002] Various noises may occur during a telephone conference or a video conference. Some of the noises may be impulsive noises, such as ticks or pops having very short duration. Other noises may be constant noises, such as the sound from an airconditioning unit. Conference participants may also create various percusive noises by typing on a computer keyboard, eating, shuffling papers, whispering, tapping a table with a pen, or the like. Like the other forms of noise, the percusive noises can be picked up by the conferencing equipment and sent to far end participants who may find the noises distracting or disruptive. However, the percusive noises are different than some of the other forms of noise, such as constant noises, that may occur during a conference because of the frequency range where percussive noises are in relation to the frequency range for speech. Noise reduction techniques known in the art are generally capable of handling constant noises during a conference. However, current noise reduction techniques are not well suited to handle percussive noises. Therefore, a need exits for a method and apparatus for reducing percussive noises in a conference.

[0003] A conferencing unit reduces or eliminates percussive noise that it outputs to a far-end. When receiving near-end audio from a microphone, the unit filters the near-end audio into a plurality of bands and thresholds the instantaneous energy relative to background noise energy for each band. The thresholded energies in a first range of bands are summed, as are the thresholded energies in a second range of bands. The first range is preferably 300 to 600-Hz, while the second range is preferably 100 to 300-Hz and 600-Hz to 14-kHz. Using a calculation, these two sums are compared to a threshold value to determine whether speech is present in the near-end audio received. Based on the determination, the conferencing unit adjusts an output level of the near-end audio to reduce or eliminate any percussive noise in audio that it outputs to far-end units. Preferably, the threshold value used in the comparison depends on whether there is far-end audio present in the near-end audio currently being received by the unit for processing.

[0004] The foregoing summary is not intended to summarize each potential embodiment or every aspect of the present disclosure.

[0007] Referring to FIG. 1, a conferencing system 10 is illustrated. The conferencing system 10 can be a telephone conferencing system, a videoconferencing system, a desktop conferencing system, or other type known in the art. During the conference, a near-end participant 20 may speak from time to time. All the while, an audio input module 110 of a near-end conferencing unit 100 receives near-end audio from a microphone 112. The unit 100 processes the near-end audio and sends output audio via a network interface 170 to one or more far-end conferencing units 30 where a far-end participant 40 can hear the output audio. Likewise, the network interface 170 receives far-end audio from the far-end unit 30. The conferencing unit 100 processes the far-end audio and sends it to a speaker 122 via an audio output module 120 for the near-end participant 20 to hear.

[0008] In addition to speaking, some form of percussive noise, such as typing sounds on a keyboard of a computer 22, may be generated during the conference. If the percussive noise is sent as is to the far-end unit 30, the far-end participant 40 may find the it disruptive or distracting. To reduce the effects of percussive noise, the conferencing unit 100 includes a percussive noise detector 150 and an auto-mute function 160. These components are used in conjunction with some conventional components of the conferencing unit 100, which can be, for example, a Polycom HDX series videoconferencing unit.

[0009] The conventional components of the conferencing unit 100 include a band pass filter bank 130, a control module 140, a network interface 170, and an echo-canceling module 180. Because these conventional components are known in the art, they are not described in detail here. In general, these components 130, 140, 170, and 180 can be discrete components or can be integrated together. The filter bank 130, control module 140, and echo-cancellation module 180 can be comprised of one or more of a microcontroller, programmable Digital Signal Processor, Field Programmable Gate Array, or application-specific integrated circuit. The network interface 170 can be comprised of any conventional interfaces for teleconferencing and videoconferencing.

[0010] How the conferencing unit 100 reduces the effects of percussive noise using the percussive noise detector 150 and auto-mute function 160 will now be explained with reference to one example of a percussive noise reduction process 200 in FIG. 2. In general, this process 200 operates as a sensitive speech detector in software and hardware components of the conferencing unit 100 and detects speech while not being overly effected by percussive noises. When speech is not detected but percussive noise is present, the process 200 reduces or mutes the audio gain that it outputs to the far end. In this way, when there is no speech in the near-end audio, any irritating percussive noises can be reduced or eliminated from output audio being sent to the far-end unit 30. In one example, the percusive noise reduction process 200 may be configured to reduce non-speech, percussive noises, like typing sounds, by 14 dB or even more (for instance, a complete mute) when there is substantially no speech present in the near-end audio.

[0011] Initially in the process 200, the conferencing unit 100 recieves near-end audio from the microphone 112 (Block 205). This near-end audio may or may not include speech and percussive noises. The band pass filter bank 130 separates the received audio into a plurality of bands spanning a desired frequency range (Block 205). In one embodiment, for example, 640 bands spanning a frequency range from 0 to 16-kHz are used so that each band has a width of 25-Hz. These filtered bands will then be used in subsequent processing to reduce percusive noises, but as one skilled in the art will appreciate, there may be a number of other uses for the filtered bands from the bandpass filter bank 130.

[0012] After separating the audio into bands, the process 200 then enters a looped operation from Block 215 to Decision 240 to find the instantaneous energy for each band. Preferably, the absolute value for the instantaneous energies is used. In particular, a given band is selected (Block 205), and its absolute instantaneous energy is determined for a given time period (Block 220). Then, an estimate is calculated for the energy of the background noise in the band using known techniques for background noise estimation (Block 205). In general, the background noise estimation maintains a running minimum energy that has occurred in a previous period of time and uses this minimum as an estimate of the background noise.

[0013] An energy ratio of the instantaneous energy with respect to the background noise energy for the given band is calculated (Block 230). If the calculated energy ratio for this band is greater than a threshold (e.g., T_energy=40), then the instantaneous energy for the band is left alone. Otherwise, the value of the instantaneous energy for the band is set to zero if the ratio falls below the threshold (Block 235). Zeroing out the energy's value essentially eliminates this band from further consideration in later processing, such as when the energies are summed together as discussed below. Finally, the looped operation is repeated for another of the bands (See Decision 240).

[0014] Once the instantaneous energies have been calculated for each of the bands, the process 200 finds a first sum of the energies for bands lying within a mid-band range (Block 245) and finds a second sum of the energies for bands lying in an outer-band range (Block 250). In one embodiment, the mid-band range for the first sum encompasses a frequency range between about 300-Hz and about 600-Hz, and the outer-band range for the second sum encompasses a frequency range of about 100 to 300 Hz and about 600 Hz to 14 kHz. Experimentation has found that having these ranges is preferred, because speech may occur during a conference primarily in the preferred mid-band range and speech can be suitably differentiated from percussive noises using these ranges. However, other values for the ranges could be used. For example, the lower end of the mid-band range could be as low as 10-Hz instead of 300-Hz, or the upper end of the range could be as high as 6000-Hz instead of 600-Hz with the limits of the outer-band ranges adjusted accordingly.

[0015] The process 200 then compares the first and second sums to determine whether speech is present in the near-end audio being currently processed (Block 255). Preferably, speech is declared present if the following calculation is satisfied: [(First Sum)−A*(Second Sum)]>T_speech. Thus, speech is present if the difference of the first sum of the energies in the mid-band range less the second sum of energies in the outer-band range multiplied by a weighting factor A is greater than a threshold value, T-_speech. If speech is not substantially present in the near-end audio received, then the auto-mute function 160 is activated to either mute or reduce the gain of the audio communicated to the network interface 170 (Block 260). In this way, the auto-mute function 160 acts to eliminate or reduce the amount of near-end percussive noise that will be present in the audio output to the far-end unit 30.

[0016] In the calculation at Block 255, the value of the weighting factor A is a constant set at 16. The value used for the threshold, T_speech, however, depends on whether the loudspeaker 122 is active as determined by an input from the echo-canceling module 180 of FIG. 1. As shown in FIG. 1, the echo-canceling module 180 receives audio from the network interface 170 that has come from the far-end unit 30. Before sending the audio to the output module 120 for delivery by the speaker 122, the echo-canceling module 180 performs conventional echo-cancellation processing on the audio using techniques known in the art to reduce any echo caused by the microphone 112 receiving feedback from the speaker 122 when active.

[0017] In addition to conventional echo cancellation, the echo-canceling module 180 sends an input signal to the percussive noise detector 150 to indicate whether the loudspeaker 122 is active or not. If the input signal indicates that the loudspeaker 122 is inactive (i.e., only near-end audio is currently or primarily being received), then the percussive noise detector 150 gives the threshold, T_speech, a lower value of about 20,000 in the calculation of Block 255. If the input signal indicates that the loudspeaker 122 is active (i.e., the near-end audio received includes far-end audio output from the speaker 122), then the detector 150 gives T_speech a higher value of about 400,000.

[0018] The threshold T_speech and the weighting factor A used in the calculation of Block 255 have been experimentally found to help separate speech from potential percussive noise and to help eventually reduce or eliminate percussive noise when speech is substantially absent from the near-end audio received. Although the previous example has given only two values for the threshold T_speech, the threshold can have more than two valuations depending on the implementation and can be adjusted between several valuations in relation to the amount of echo-cancellation used and/or far-end audio signal present in the active loudspeaker 122.

[0019] Once auto-mute has been implemented as needed and the audio has been output to the network interface 170, the process 200 returns to Block 205 and repeats processing near-end audio for a new time interval, which can be every 20-ms or so. As the conference progresses, the conferencing unit 100 may mute or reduce output audio from time to time by various amounts depending on whether speech is present in the near-end audio and whether the loudspeaker 122 is active. In this way, any percussive noises that occur during the conference can be reduced or eliminated when the near-end participant 20 is not speaking. This is then intended to reduce the amount of disruptive percussive noise sent to the far-end unit 30.

[0020] The percussive noise reduction process 200 of FIG. 2 can be ultimately coded into a computer code and stored on a computer-readable media, such as a compact disk, a tape, stored in a volatile or non-volatile memory, etc. Accordingly, the percussive noise reduction process 200 of FIG. 2 can comprise instructions stored on a program storage device for causing a programmable control device to perform the process. Using an interface (e.g., RS-232 Serial Interface) between a computer and the conferencing unit 100 of FIG. 1, the conferencing unit 100 can also be programmed with coded software.

[0021] The foregoing description of preferred and other embodiments is not intended to limit or restrict the scope or applicability of the inventive concepts conceived of by the Applicant. In exchange for disclosing the inventive concepts contained herein, the Applicant desires all patent rights afforded by the appended claims. Therefore, it is intended that the appended claims include all modifications and alterations to the full extent that they come within the scope of the following claims or the equivalents thereof.