Baby Cry Recognizer

The present invention relates to voice recognition. More particularly, the present invention relates to recognizing particular sounds such as the crying sound of a baby.

Voice recognition relates to detecting and discriminating voice and speech sounds. Voice recognition may be used in devices which are triggered by voice or speech sounds such as a baby cry recognizer. Several methods have been proposed for voice recognition including a filter bank method and a method which uses linear prediction codes. Both of these methods have a good recognition rate but are complicated and require expensive hardware.

Another, less complicated, technique utilizes the zero crossing rate (ZCR) feature of the inputted sound signal. FIG. 1 shows a graph of an exemplary inputted sound signal. As depicted, the signal traces an undulating path above and below a zero polarity axis which axis is located at the center between the maximum and minimum signal levels of the inputted sound signal. The ZCR is simply the rate in which the path traced by the sound signal crosses the zero polarity axis (i.e., the number of zero crossings in a frame or time interval divided by the duration of that frame or time interval). Generally, ZCR techniques have a substantially lower recognition rate than the two aforementioned techniques.

Several ZCR devices have been proposed for voice recognition. R.O.C. Patent Nos. 28117 and 40308, propose systems in which sound signals both with and without voice sounds are analyzed. Thereafter, a comparison, such as a distance measurement, is performed on sound signals containing voice sounds. Such systems are complex as both voiceless and voiced sound signals must be analyzed.

U.S. Pat. Nos. 4,763,278 and 4,780,906, disclose speech recognizers which convert the ZCR of an inputted sound signal into a feature vector containing information about the ZCR and energy of the inputted sound signal. The feature vector is compared to predetermined template vectors. The recognized word may then be selected as the word having a template vector which best matches the feature vector of the inputted sound signal. A device according to these patents, however, is complex as it requires the computation of a feature vector.

U.S. Pat. No. 4,799,171, discloses a device which uses the ZCR of an inputted sound signal to determine whether the frequency of the sound signal is in a low frequency band (indicating that the sound signal contains the word "yes") or a high frequency band (indicating that the sound signal contains the word "no"). Such a device does not provide an acceptable recognition rate and is thus used only for toys.

It is an object of the present invention to overcome the disadvantages of the prior art devices.

This and other objects are achieved by the present invention which provides a device and process for recognizing a particular sound, such as a baby's cry, in an inputted sound signal if two conditions are met:

1. more than a first threshold C.sub.1 of frames of a preselected number of frames C in a predetermined length time interval satisfy f.sub.1 .ltoreq.ZCR.sub.f .ltoreq.f.sub.2 where ZCR.sub.f is the zero crossing rate of the sound signal during a frame and f.sub.1 and f.sub.2 are lower and upper frequencies of a particular frequency range, and

2. the zero crossing rate during the entire time interval satisfies C.sub.2 .ltoreq.ZCR.sub.T .ltoreq.C.sub.3 where C.sub.2 is a low frequency threshold, C.sub.3 is a high frequency threshold and ZCR.sub.T is the zero crossing rate during the entire time interval.

A baby cry recognizer according to one embodiment of the present invention includes an amplifier circuit which amplifies a received sound signal. In response to the amplified sound signal, a pulse generator circuit generates a pulse signal with zero crossings which are aligned with the zero crossings of the amplified sound signal. The pulse signal, in turn, is inputted to a signal recognition circuit. The signal recognition circuit, obtains the zero crossing rate of the pulse signal during each of a preselected number of frames in a predetermined length time interval. The signal recognition circuit then determines if the zero crossing rate of the pulse signal is within a particular frequency range during more than a first threshold number of frames of the preselected number of frames. The signal recognition circuit also determines if the zero crossing rate of the pulse signal, over the entire time interval, is greater than or equal to a second low threshold but less than or equal to a third high threshold. If these two conditions are satisfied, then the signal recognition circuit outputs a signal indicating that a baby's cry was detected.

Illustratively, prior to determining whether the two above mentioned conditions are satisfied, the signal recognizer first evaluates a precondition to determine if a baby's cry was potentially received. In evaluating the precondition, the signal recognizer determines the zero crossing rate in each frame of a predetermined length time interval. Then, the signal processor determines if the zero crossing rate of the pulse signal during each frame of the time interval is at least equal to a fourth threshold. If so, the signal recognizer evaluates the aforementioned conditions for determining if a baby's cry was detected. Otherwise, the signal recognizer evaluates the same precondition for the next succeeding time interval.

In short, a baby cry recognizer is provided which is simple and inexpensive yet has a reasonably good accuracy. Since the baby cry recognizer uses ranges and thresholds which may be predetermined, it need not be trained to the voice of the particular subject to be monitored.

FIG. 1 graphically depicts an exemplary sound signal waveform and its zero crossings.

FIG. 2 depicts a block diagram of a baby cry recognizer according to an embodiment of the present invention.

FIG. 3 depicts an exemplary amplified signal waveform and a corresponding pulse signal waveform produced by the circuit of FIG. 2.

FIG. 4 depicts a more detailed circuit according to an embodiment of the present invention.

FIG. 5 schematically depicts a flow chart illustrating the process executed by the signal recognizer of FIGS. 2 and 4.

Referring now to FIG. 2, a baby cry recognizer 10 according to the present invention is depicted in block form. The baby cry recognizer 10 includes a microphone 1 which detects sounds, including, for example, a baby's cry. This sound signal is converted to electrical form therein and inputted to an amplifier 2. Illustratively, the amplifier 2 amplifies the signal to produce a 0 to 5 volt peak to peak amplitude signal as depicted in FIG. 3. Such a signal has as 2.5 volt zero polarity axis.

Referring again to FIG. 2, the amplified signal outputted from the amplifier 2 is inputted to a pulse generator 3. As depicted in FIG. 3, the pulse generator 3 generates a pulse signal having zero crossings which are aligned with the zero crossings of the amplified signal. For example, a series of rectangular pulses may be generated wherein each zero crossing of the inputted signal is approximately aligned with a corresponding rising or falling edge of a rectangular pulse of the pulse signal. Illustratively, it is advantageous to convert the amplified sound signal to a pulse signal so as to remove other extraneous features of the signal (e.g., the signal's amplitude, shape, etc.) which are not of interest in detecting a baby's cry according to the present invention.

The pulse signal generated by the pulse generator 3 is inputted to a signal recognizer 4. Therein, the pulse signal is analyzed to determine if a baby's crying sound was received. A process for recognizing a baby's cry is discussed in detail below. The signal recognizer 4 recognizes a baby's cry if two conditions are satisfied. First, the signal recognizer 4 determines if more than a first threshold number of frames C.sub.1, of a preselected number of frames C within a time interval, satisfy f.sub.1 .ltoreq.ZCR.sub.f .ltoreq.f.sub.2, where f.sub.1 and f.sub.2 are low and high frequencies of a particular frequency range and ZCR.sub.f is the ZCR during a frame. In the above determination, the number of zero crossings in each frame may be substituted for each ZCR.sub.f in the expression f.sub.1 .ltoreq.ZCR.sub.f .ltoreq.f.sub.2 provided f.sub.1 and f.sub.2 are replaced by their corresponding equivalent number of zero crossings in a frame (i.e., f.sub.1 and f.sub.2 are each multiplied by the duration of a frame). Second, the signal recognizer 4 determines if C.sub.2 .ltoreq.ZCR.sub.T .ltoreq.C.sub.3 is satisfied where ZCR.sub.T is the ZCR over an entire time interval, C.sub.2 is a second low threshold and C.sub.3 is a third high threshold. Again, the number of zero crossings in a time interval may be substituted for ZCR.sub.T in the expression C.sub.2 .ltoreq.ZCR.sub.T .ltoreq.C.sub.3 provided C.sub.2 and C.sub.3 are replaced by their corresponding equivalent number of zero crossings in a time interval (i.e., C.sub.2 and C.sub.3 are each multiplied by the duration of a time interval).

If a baby's cry is detected (i.e., these two conditions are satisfied), the signal recognizer 4 outputs an indication signal. This indication signal may be an alarm which alerts an attendant. Alternatively, the indication signal triggers a pacifying device such as a motor which rocks the cradle in which the baby is resting.

It is not necessary to train the device 10, i.e., it is not necessary to execute a preprocess in which the monitored baby's voice is inputted to the device 10 to set parameters therein for recognizing the baby's voice. Rather, predetermined thresholds and frequency ranges are illustratively used in the signal recognizer 4 according to a process (discussed in detail below) which accurately recognizes the cry of a broad segment of the population of babies.

Turning now to FIG. 4, a more detailed circuit diagram for the baby cry detector 10 is shown. Inputted sound signals are converted by the microphone 1 into electrical signals. Illustratively, the microphone 1 is powered by a voltage source V.sub.cc via a resistor R.sub.1. The electrical signal generated by the microphone 1 is coupled through a coupling capacitor C.sub.1 to the inverting input of an op amp U.sub.10. (The capacitor C.sub.1 blocks dc signals). As depicted, the op amp U.sub.10, resistors R.sub.2 and R.sub.3 and capacitors C.sub.1 and C.sub.2 are connected to form a bandpass amplifier, with roll off points at (R.sub.2 C.sub.1).sup.-1 and (R.sub.3 C.sub.2).sup.-1. Illustratively, R.sub.2, R.sub.3, C.sub.1 and C.sub.2 are selected to filter out sounds outside of the frequency range of human voices. Furthermore, R.sub.2 and R.sub.3 are illustratively selected so as to adequately amplify the signal outputted by the microphone 1 to a 5 volt peak to peak amplitude.

It may also be desirable to ensure that the voltage level of the amplified sound signal is strictly in the range of 0 to the maximum positive peak output of the op amp U.sub.10 (e.g., 5 volts). To that end, a dc voltage may be inputted to the non-inverting input of the op amp U.sub.10 so that the output signal has a component equal to the maximum positive peak output of the op amp U.sub.10 (e.g., 5 volts). This may be achieved by connecting a voltage divider network R.sub.4 and R.sub.5 to V.sub.cc as shown (the capacitor C.sub.3 may also be added to remove any ac component of the non-inverting input signal). The two output signal components (the amplified inverted sound signal, and the amplified dc signal) are added together to produce an amplified sound signal with the appropriate peak to peak amplitude (e.g., 0 to 5 volts).

The signal outputted from the op amp U.sub.10 is fed to the pulse generator 3. Illustratively, a second op amp U.sub.20 is used in order to generate the pulse signal. To generate a rectangular pulse signal with zero crossings which are aligned with the zero crossings of the amplified signal, the op amp U.sub.20 is connected as a comparator. The amplified sound signal is inputted to the non-inverting input of the op U.sub.20. A constant dc input voltage equal to the zero polarity axis level (e.g., 2.5 volts) is illustratively supplied to the inverting input of the op amp U.sub.20. This may be achieved by inputting the maximum peak amplitude of the amplified signal (e.g., 5 volts) through a voltage divider network R.sub.6 -R.sub.7, where R.sub.6 =R.sub.7. It may be appreciated that both inputs are overdriven thereby producing a clipped or rectangular shaped output waveform. Wherever the amplitude of the amplified signal is less than the zero polarity axis, i.e., less than 2.5 volts, the dc signal of the inverting input dominates the input, thereby producing an output signal with a constant minimum level. Wherever, the amplified signal exceeds the zero polarity axis, however, the amplified sound signal of the non-inverting input dominates the input thereby producing an output signal with a constant maximum level.

The pulse signal thus generated is inputted to a signal recognizer circuit 4, such as the microcomputer U.sub.30. Illustratively, the pulse signal is inputted to the lead T1 of an Intel 8048 single chip microprocessor. An indication signal is outputted from lead P0 if a baby's cry is recognized.

Turning now to FIG. 5, the determination process performed by the signal recognizer circuit 4 (e.g., the microcomputer U.sub.30) is now discussed. In a first step 102, the signal recognizer performs various initialization procedures such as resetting any parameters and or circuitry (e.g., counters) which may be utilized therein. Next in step 104, the signal recognizer circuit determines the ZCR for every frame (referred to as ZCR.sub.f) in a predetermined time interval. Illustratively, the predetermined time interval is 1 second and the duration of each frame is 10 milliseconds. In step 106, the signal recognizer circuit determines if, from the ZCR.sub.f measured in each frame of the predetermined time interval, a crying sound may have been transmitted. For example, the signal recognizer may determine if the ZCR.sub.f in each frame is greater than or equal to a threshold frequency, e.g., 1600 Hz. If this is not the case, the process loops back to step 104 wherein the ZCR.sub.f is again measured in each frame (10 msec) of the next, successive time interval (1 sec). If, on the other hand, the ZCR.sub.f in each frame does exceed the threshold (e.g., 1600 Hz), the signal recognizer proceeds to step 108.

Alternatively, the signal recognizer may determine the number of zero crossings in each frame (10 msec) of the time interval (1 sec) in step 104. In such a case, the signal recognizer determines if the number of zero crossings in each frame is greater than or equal to a threshold number of zero crossings, e.g., 16, in step 106.

In step 108, the ZCR.sub.f is obtained for a preselected number of frames C within the predetermined time interval. Illustratively, C equals 100 frames. (Thus, the ZCR.sub.f is obtained for each frame in an entire time interval of 1 sec.) Next in step 110, the signal recognizer determines if the number of these C frames in which the ZCR.sub.f falls in the range f.sub.1 .ltoreq.ZCR.sub.f .ltoreq.f.sub.2 is greater than the threshold C.sub.1. Illustratively, f.sub.1 equals 1600 Hz, f.sub.2 equals 3200 Hz and C.sub.1 equals 10 frames. Alternatively, the number of zero crossings is obtained in C frames in step 108. In such a case, then, in the above example, f.sub.1 illustratively equals 16 crossings and f.sub.2 equals 32 crossings. If the aforementioned condition of step 110 is not satisfied, the signal recognizer returns to step 102. If this condition is satisfied, the signal recognizer performs step 112.

In step 112, the signal recognizer determines whether the ZCR over the entire time interval (referred to as ZCR.sub.T) is between a low threshold C.sub.2 and high threshold C.sub.3, i.e., C.sub.2 ZCR.sub.T .ltoreq.C.sub.3. Illustratively, C.sub.2 equals 1600 Hz and C.sub.3 equals 3200 Hz. Because ZCR.sub.T is the ZCR over the entire time interval of 1 sec the thresholds of C.sub.2 =1600 zero crossings and C.sub.3 =3200 zero crossings may be used. If this condition is not satisfied, the signal recognizer returns to step 102. However, if this condition is satisfied, the signal recognizer outputs a signal indicating that a baby's cry was received in step 114. Such a signal may be used to trigger an alarm to alert an attendant or may be used to trigger a pacifying device, e.g., a motor which rocks the cradle in which the baby is resting.

In summary, a baby cry detector is disclosed which is both simple and inexpensive. The invention analyzes the zero crossing rate of an inputted sound signal using simple threshold conditions. Despite the simplicity of the detector, a baby's cry may be recognized with reasonably good accuracy. Moreover, a detector according to the present invention need not be trained to recognize the voice of the particular monitored subject in order to recognize a baby's cry.

Finally, the aforementioned embodiments are intended to be merely illustrative. Numerous other embodiments may be devised by those having ordinary skill in the art without departing from the spirit and scope of the following claims.