JP2012163682A - Voice processor and voice processing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a voice processor capable of enhancing articulation of receiving voice even when transmission by a user himself or herself is in progress, and further to provide a voice processing method.SOLUTION: A voice signal input from a microphone is divided into a plurality of voice signals of predetermined frequency bands, weighting processing according to a signal-to-noise ratio is executed with respect to a voice signal of each frequency band and an amount of back ground noise that is included in the voice signal of each frequency band having been divided is estimated. Further, a receiving voice signal is divided into a plurality of voice signals of each predetermined frequency band, a gain that is applied to the voice signals of each frequency band is set based on the amount of the back ground noise of the each frequency band having been estimated such that the voice signals of the each frequency band are increased as the amount of the back ground noise is increased, and the receiving voice signal is corrected by multiplying the voice signals of the each frequency band by the gain.

Description

  The present invention relates to an audio processing apparatus and method for correcting an audio signal according to background noise.

  For example, since a mobile phone is often used outdoors or the like, it is often difficult to hear a received voice due to background noise such as ambient traffic noise and advertisement noise. This is because the intelligibility of the received voice is lowered by masking a part of the frequency components of the received voice by the background noise. Therefore, various techniques have been proposed for correcting a voice signal (received voice) according to background noise so that the received voice can be easily heard even under noise.

For example, in Patent Document 1, the clarity of received speech is improved by shaping the speech spectrum of the received speech into a target spectrum (for example, a speech spectrum in which a frequency component having a relatively low signal level such as a consonant is increased). The technology is described. Furthermore, in Patent Document 1, in order to prevent the transmitted voice due to the user's own utterance as background noise and prevent the received voice from being extremely amplified, the voice spectrum of the voice signal input from the microphone is analyzed. It is determined whether or not the voice signal is the user's own transmission voice, and when it is determined that the voice signal is the user's own transmission voice, the following three types of processing (1) to (3) are performed. It is described to be executed.
(1) In the period determined to be the user's own transmitted voice, the filter coefficient of the filter unit for shaping the voice spectrum of the received voice into the target spectrum is an initial value (for example, a value for outputting the received voice as it is) ) (Hereinafter referred to as first background processing).
(2) The filter coefficient is suppressed to a preset maximum value or less during a period determined as the user's own transmitted voice (hereinafter referred to as second background processing).
(3) The update of the filter coefficient is stopped in a period when it is determined that the user's own voice is transmitted. That is, the filter coefficient immediately before the determination as the user's own transmitted voice is used (hereinafter referred to as third background processing).

JP 2004-061617 A

  The technique described in Patent Document 1 described above has a problem that the intelligibility of the received voice cannot be improved when the user is speaking.

  For example, in the first background processing, while the user himself / herself is speaking, the received voice is not corrected by fixing the filter coefficient with, for example, an initial value. Therefore, the clarity of the received voice cannot be improved while the user himself is speaking.

  In the second background processing, the received voice is corrected even when the user is speaking, but the received voice is not necessarily shaped into the target spectrum by limiting the filter coefficient to the maximum value or less. Absent. Therefore, the clarity of the received voice may not be improved while the user himself is speaking.

  Further, in the third background processing, in order to maintain the state immediately before it is determined as the user's own transmitted voice (the value of the filter coefficient), the user's own voice follows the change in the background noise during the utterance. The sound cannot be corrected. Therefore, as in the second background process, the clarity of the received voice may not be improved while the user himself / herself is speaking.

  The present invention has been made in order to solve the problems of the background art as described above, and is a speech processing apparatus capable of improving the intelligibility of received speech even during the transmission of the user himself / herself. And to provide a method.

In order to achieve the above object, an audio processing device of the present invention includes a first frequency analysis unit that divides an audio signal input from a microphone into a plurality of first audio signals in a predetermined frequency band;
The first audio signal for each frequency band divided by the first frequency analysis unit is weighted according to the signal-to-noise ratio, and the background noise amount for each frequency band included in the audio signal is estimated. A background noise estimation unit;
A second frequency analyzer that divides the received voice signal into a plurality of second voice signals for each predetermined frequency band;
Based on the background noise amount for each frequency band estimated by the background noise estimation unit, the frequency is set so that the second sound signal output from the second frequency analysis unit increases as the background noise amount increases. Setting a gain to be applied to the second audio signal for each band, and multiplying the corresponding second audio signal of the frequency band by the gain to correct the received audio signal;
A reception voice correction unit including a frequency synthesis unit that synthesizes the frequency of the corrected second voice signal for each frequency band output from the characteristic correction unit, and reproduces the corrected reception voice signal;
Have

On the other hand, the audio processing method of the present invention divides an audio signal input from a microphone into a plurality of first audio signals in a predetermined frequency band, and has a signal-to-noise ratio with respect to the first audio signal for each frequency band. Performing a corresponding weighting, estimating a background noise amount for each of the divided frequency bands included in the audio signal,
The received voice signal is divided into a plurality of second voice signals for each predetermined frequency band, and based on the estimated background noise quantity for each frequency band, the larger the background noise quantity, the corresponding second voice signal becomes. Set a gain to be applied to the second audio signal for each frequency band so as to increase,
The received audio signal is corrected by multiplying the corresponding second audio signal of the frequency band by the gain,
This is a method of frequency-synthesizing the second audio signal for each frequency band after the correction and reproducing the corrected received voice signal.

  According to the present invention, it is possible to improve the intelligibility of received voice even during the transmission of the user himself / herself.

It is a block diagram which shows the example of 1 structure of the speech processing unit of 1st Embodiment. It is a block diagram which shows the example of 1 structure of the background noise estimation part shown in FIG. It is a block diagram which shows one structural example of the characteristic correction | amendment part of 1st Embodiment. It is a block diagram which shows the example of 1 structure of the speech processing unit of 2nd Embodiment. It is a block diagram which shows the example of 1 structure of the characteristic correction | amendment part of 2nd Embodiment. It is a schematic diagram which shows the process example of the gain limiter part shown in FIG. It is a block diagram which shows the example of 1 structure of the characteristic correction | amendment part of 3rd Embodiment.

Next, the present invention will be described with reference to the drawings.
(First embodiment)
FIG. 1 is a block diagram illustrating a configuration example of the speech processing apparatus according to the first embodiment.

  As shown in FIG. 1, the speech processing apparatus according to the first embodiment is configured to include a transmitted speech analysis unit 1 and a received speech correction unit 2.

  The transmitted voice analysis unit 1 includes a first frequency analysis unit 10 and a background noise estimation unit 11. The received voice correction unit 2 includes a second frequency analysis unit 12, a characteristic correction unit 13, and a frequency synthesis unit 14.

  The audio processing device shown in FIG. 1 is, for example, an A / D converter that converts an audio signal (analog signal) input from a microphone into a digital signal, and an audio signal (digital signal) output from the received audio correction unit 2. D / A converter for converting to analog signal, CPU for processing audio signal according to program, DSP for executing arithmetic processing, memory for storing various data necessary for program and processing, various logic circuits, etc. Including a known signal processing circuit.

  The first frequency analysis unit 10 divides the audio signal input from the microphone into a plurality of frequency band audio signals (first audio signals) by processing such as DFT (Discrete Fourier Transform). The first frequency analysis unit 10 may be realized by a band division filter such as an IIR (Infinite Impulse Response) filter. In addition, the first frequency analysis unit 10 may divide the audio signal input from the microphone with a certain bandwidth, and in consideration of human auditory characteristics, for example, the higher frequency may be divided so that the bandwidth becomes wider. May be.

  The background noise estimation unit 11 estimates the amount of background noise for each frequency band divided by the first frequency analysis unit 10 included in the audio signal input from the microphone. At this time, if the user's own transmission voice signal (hereinafter referred to as a transmission voice signal) is included in the voice signal, the remaining background noise amount is calculated by removing the transmission voice signal as much as possible. .

  FIG. 2 is a block diagram illustrating a configuration example of the background noise estimation unit illustrated in FIG.

  As shown in FIG. 2, the background noise estimation unit 11 includes a weighted speech calculation unit 100 and an estimated noise calculation unit 101.

  The weighted voice calculation unit 100 calculates the background noise and the background noise amount for each frequency band estimated by the estimated noise calculation unit 101 and the voice signal for each frequency band input from the microphone including the user's own voice. The SNR (Signal-to-Noise Ratio) for each frequency band is calculated and weighted according to the SNR for the audio signal for each frequency band input from the microphone.

  The estimated noise calculation unit 101 estimates the background noise amount for each frequency band from the audio signal for each frequency band input from the microphone and the weighted audio signal for each frequency band output from the weighted audio calculation unit 100. Based on the weighted sound signal for each frequency band output from the weighted sound calculation unit 100, the estimated noise calculation unit 101 calculates the background noise amount for each frequency band from the sound signals of a predetermined number of frames estimated as background noise. While calculating, the value is updated.

  A specific method of calculating the background noise amount by the background noise estimation unit 11 shown in FIG. 2 is described in, for example, Japanese Patent Application Laid-Open No. 2008-216721.

  In this way, the audio signal for each frequency band input from the microphone used for updating the estimated background noise amount is weighted according to the SNR, and the background noise amount is estimated, so that the input from the microphone is performed. Since the influence of the user's own transmission voice included in the received voice signal can be reduced, the background noise amount can be estimated more accurately.

  The second frequency analysis unit 12 converts a received voice signal (hereinafter referred to as a received voice signal) into a plurality of frequency band voice signals (second voice signal) in the same manner as the first frequency analysis unit 10 by processing such as DFT. Divide into

  The characteristic correction unit 13 can be realized, for example, with the configuration shown in FIG.

  FIG. 3 is a block diagram illustrating a configuration example of the characteristic correction unit according to the first embodiment.

  As illustrated in FIG. 3, the characteristic correction unit 13 includes a smoothing unit 200, a gain generation unit 201, a gain Matrix unit 202, and a correction unit 203.

  The smoothing unit 200 smoothes the background noise amount for each frequency band estimated by the background noise estimation unit 11 in the time axis direction or the frequency axis direction. The smoothing unit 200 may smooth the background noise amount in each frequency band in both the time axis direction and the frequency axis direction. For example, when the speech processing apparatus executes processing on speech signals in units of predetermined frames, the background noise amount for each of a plurality of frames adjacent in the time axis direction may be smoothed. Alternatively, the background noise amount may be smoothed for each of a plurality of frames adjacent in the frequency axis direction. Alternatively, the background noise amount for each of a plurality of frames adjacent in the time axis direction and the frequency axis direction may be smoothed. By providing the smoothing unit 200, it is possible to suppress the received voice from unnaturally changing even when the background noise amount changes abruptly.

  The gain generation unit 201 calculates an amplification factor (gain) to be applied to the received voice signal for each frequency band from the background noise amount of each frequency band smoothed by the smoothing unit 200. The gain generation unit 201 may set the gain of the received voice signal for each frequency band so that the gain increases as the background noise amount increases, for example. The gain generation unit 201 may obtain the gain of the received voice signal for each frequency band by, for example, a linear expression of the background noise amount or a higher-order equation of the second or higher order.

  The gain Matrix unit 202 mixes and smoothes the gain of each frequency band calculated by the gain generation unit 201 using, for example, the following equation (1). Formula (1) shows an example of a calculation formula by the gain Matrix unit 202 when the received voice signal is divided into N (N is a positive number) frequency bands (f1, f2,..., FN from the lowest frequency). ing. The left side of Expression (1) is the gain for each frequency band (f1, f2,..., FN) after processing by the gain Matrix unit 202.

  The gain Matrix unit 202 smoothes the gain for each frequency band calculated by the gain generation unit 201 in the frequency axis direction. The mixing coefficient is basically set so that the value increases in the adjacent frequency band. Each mixing coefficient may be set in advance, or may be dynamically determined from the distribution status of each gain calculated by the gain generation unit 201. For example, when the variance of each gain is large, smoothing may be performed using gains in more frequency bands. By performing such processing by the gain Matrix unit 202, it is possible to prevent the received voice after frequency synthesis by the frequency synthesis unit 14 from becoming unnatural. Note that it is not necessary to set all values for the mixing coefficient, and some of the mixing coefficients can be set to “0”. In that case, the more the mixing coefficient set to “0”, the more the amount of calculation can be reduced, and the memory holding the mixing coefficient can be saved. However, the mixing coefficient needs to be set so that the gain of each frequency band after the processing by the gain Matrix unit 202 does not become “0”. For example, by setting the mixing coefficient corresponding to each frequency band to “1” and the other mixing coefficients to “0”, the gain of each frequency band calculated by the gain generation unit 201 is output as the processed gain as it is. It is possible to make it.

  The correction unit 203 multiplies the audio signal of the corresponding frequency band by the gain for each frequency band after processing output from the gain Matrix unit 202 and outputs the result.

  The frequency synthesizer 14 performs inverse transformation of processing (DFT or the like) by the second frequency analyzer 12 to synthesize the frequency of the audio signal for each frequency band output from the characteristic corrector 13 and receive the received audio signal ( Play (after correction).

According to the audio processing apparatus of the first embodiment, the audio signal input from the microphone is weighted according to the SNR for each frequency band, and the background noise amount is estimated, so that the audio signal is input from the microphone. Since the influence of the user's own transmitted voice included in the voice signal can be reduced, the background noise amount can be estimated more accurately. Therefore, in the estimated background noise amount, the influence of the user's own transmitted voice is reduced. Therefore, if the received voice is corrected based on the background noise amount, the user's own speech is being transmitted. Also, the received voice can be corrected, and the clarity of the received voice can be improved.
(Second Embodiment)
FIG. 4 is a block diagram illustrating a configuration example of the speech processing apparatus according to the second embodiment, and FIG. 5 is a block diagram illustrating a configuration example of the characteristic correction unit according to the second embodiment.

  As shown in FIG. 4, the speech processing apparatus according to the second embodiment has a configuration including a characteristic correction unit 15 instead of the characteristic correction unit 13 shown in the first embodiment. As shown in FIG. 5, the characteristic correction unit 15 of the second embodiment has a configuration in which a gain limiter unit 204 is added to the characteristic correction unit 13 shown in the first embodiment. Since other configurations and operations of the voice processing apparatus are the same as those of the voice processing apparatus according to the first embodiment, description thereof is omitted.

  In the characteristic correction unit 13 shown in the first embodiment, since the gain calculated by the gain generation unit 201 is used as it is, a digital clip can be generated in the received voice signal after frequency synthesis by the frequency synthesis unit 14. There is sex. The gain limiter unit 204 suppresses the gain for each frequency band calculated by the gain generation unit 201, and suppresses the occurrence of a digital clip in the received voice signal after frequency synthesis.

  FIG. 6 is a schematic diagram illustrating a processing example of the gain limiter unit illustrated in FIG. The horizontal axis in FIG. 6 represents the time axis (frame), and the vertical axis represents the upper limit value (gain limit value) of the gain for each frequency band set by the gain limiter unit 204.

  First, the gain limiter unit 204 calculates the allowable maximum gain value (solid line) for each frame in which the audio signal is processed from the amplitude (digital value) of the received audio signal. The maximum gain value is obtained by dividing the maximum value that can be handled by digital signal processing by the maximum value (absolute value) of the received voice signal in the frame. The maximum gain value increases as the amplitude of the received voice signal decreases, and decreases as the amplitude of the received voice signal increases.

  The gain limiter unit 204 sets the gain limiter value (dotted line) to be equal to or less than the calculated maximum gain value. Also, the gain limiter unit 204 does not change the gain limiter value during a preset number of frames (the number of Hold frames) when the maximum gain value increases rapidly. For example, after the time T1 shown in FIG. 6, the value of the maximum gain increases, so that the gain for each frequency band can be increased. However, the gain limiter unit 204 does not increase the gain limiter value during the period of the number of Hold frames. This is because when the gain limiter value is suddenly increased and the gain for each frequency band is also suddenly increased, the volume of the received voice after frequency synthesis becomes unnatural, so that such an unnatural increase in volume is prevented. Because.

  The gain limiter unit 204 gradually increases the gain limiter value after the period of the number of Hold frames has elapsed (time T2 in FIG. 6). At this time, the gain limiter unit 204 may increase the gain limiter value at a preset ratio, or may change the increase ratio according to the background noise amount or the maximum gain value.

  When the increased gain limiter value becomes equal to the value of the maximum gain (time point T3 in FIG. 6), and when the maximum gain value increases thereafter, the gain limiter unit 204 performs the same processing as that after T1 described above. Execute. When the maximum gain value decreases, the gain limiter value is decreased in accordance with the maximum gain value.

According to the speech processing apparatus of the second embodiment, the gain limiter 204 limits the gain of each frequency band of the received speech to an upper limit value or less according to the amplitude of the received speech signal. Does not exceed the maximum value that can be handled by digital signal processing. Therefore, compared with the speech processing apparatus according to the first embodiment, the received speech can be clarified without causing deterioration in sound quality.
(Third embodiment)
FIG. 7 is a block diagram illustrating a configuration example of the characteristic correction unit according to the third embodiment.

  As shown in FIG. 7, the speech processing apparatus according to the third embodiment adds a background noise amount masking calculation unit 205 and a received speech masking calculation unit 206 to the characteristic correction unit 13 shown in the first embodiment. This is the configuration. Since other configurations and operations of the voice processing apparatus are the same as those of the voice processing apparatus according to the first embodiment, description thereof is omitted.

  The characteristic correcting unit of the present embodiment increases only the gain in the frequency band of the received voice signal masked by the background noise from the background noise amount estimated by the background noise estimating unit 11.

  The background noise amount masking calculation unit 205 calculates a known masking threshold value for each frequency band divided from the background noise amount smoothed by the smoothing unit 200, and uses the masking threshold value for each frequency band that can be heard by a person. The amount of background noise is calculated.

  The received voice masking calculation unit 206 calculates a known masking threshold value for each frequency band divided from the received voice signal, and uses the masking threshold value to calculate the received voice amount for each frequency band that can be heard by a person.

  The masking threshold refers to a sound pressure level at which a desired sound can be heard when there is another sound that masks the desired sound. Usually, since the received voice and background noise contain various frequency components, a certain frequency component of the received voice may mask other frequency components of the received voice even in the received voice. The background noise amount masking calculation unit 205 compares the masking threshold value with the background noise amount for each frequency band, and calculates the background noise amount for each frequency band that is not masked by other frequency components in the background noise and can be heard by humans. To do. Similarly, the received voice masking calculation unit 206 compares the masking threshold value with the received voice volume for each frequency band, and is not masked with other frequency components in the received voice volume, and the received voice for each frequency band that can be heard by a person. Calculate the amount. Note that a masking threshold calculation method, an audio signal correction method using a masking threshold, and the like are disclosed in, for example, Japanese Patent Application Laid-Open No. 2009-175420.

  The gain generation unit 207 compares the background noise amount output from the background noise amount masking calculation unit 205 with the received voice amount output from the received voice masking calculation unit 206 for each frequency band, and determines the background more than the received voice amount. When the amount of noise is large, the gain of the audio signal (second audio signal) in the corresponding frequency band is set to “1” or more.

  Here, if the gain of the frequency band in which the background noise amount is larger than the received voice amount is set to the background noise amount / received voice amount, the background noise amount in the frequency band and the received voice amount after the correction processing are equivalent. .

  In the frequency band where the background noise amount is larger than the received voice amount, when it is desired to always increase the received voice amount after the correction processing above the background noise amount, the gain generation unit 207 gains the voice signal (received voice) in the frequency band. May be set to a value equal to or greater than (background noise amount / received voice amount). For example, the gain of the voice signal (received voice) in the corresponding frequency band is set to (background noise amount / received voice amount) × α (α> 1.0) or (background noise amount / received voice amount) + α (α> 0.0).

  When the received voice volume is larger than the background noise level, the gain generation unit 207 sets the gain of the corresponding frequency band voice signal (received voice) to “1” so as not to be amplified.

  According to the speech processing apparatus of the third embodiment, the received speech signal is amplified only in the frequency band in which the background noise amount is larger than the received speech amount, so that the received speech does not increase unnecessarily. Therefore, the received voice having higher quality than that of the voice processing apparatus according to the first embodiment is reproduced. Note that the background noise amount masking calculation unit 205 and the received voice masking calculation unit 206 shown in FIG. 7 can be included in the characteristic correction unit 15 of the second embodiment.

DESCRIPTION OF SYMBOLS 1 Transmission voice analysis part 2 Received voice correction | amendment part 10 1st frequency analysis part 11 Background noise estimation part 12 2nd frequency analysis part 13, 15 Characteristic correction | amendment part 14 Frequency synthesis part 100 Weighted speech calculation part 101 Estimated noise calculation part 200 Smoothing unit 201 Gain generation unit 202 Gain Matrix unit 203 Correction unit 204 Gain limiter unit 205 Background noise amount masking calculation unit 206 Received voice masking calculation unit

Claims (6)

A first frequency analysis unit that divides an audio signal input from a microphone into a plurality of first audio signals in a predetermined frequency band;
The first audio signal for each frequency band divided by the first frequency analysis unit is weighted according to the signal-to-noise ratio, and the background noise amount for each frequency band included in the audio signal is estimated. A background noise estimation unit;
A second frequency analyzer that divides the received voice signal into a plurality of second voice signals for each predetermined frequency band;
Based on the background noise amount for each frequency band estimated by the background noise estimation unit, the frequency is set so that the second sound signal output from the second frequency analysis unit increases as the background noise amount increases. Setting a gain to be applied to the second audio signal for each band, and multiplying the corresponding second audio signal of the frequency band by the gain to correct the received audio signal;
A reception voice correction unit including a frequency synthesis unit that synthesizes the frequency of the corrected second voice signal for each frequency band output from the characteristic correction unit, and reproduces the corrected reception voice signal;
A speech processing apparatus. The characteristic correction unit includes:
The audio processing apparatus according to claim 1, further comprising a gain limiter unit that limits a gain for each frequency band to a predetermined upper limit value or less. The characteristic correction unit includes:
A background noise amount masking calculation unit that calculates a background noise amount for each frequency band that can be heard by a person, which is not masked by other frequency components in the background noise,
A received voice masking calculating unit that calculates a received voice amount for each frequency band that can be heard by a person, which is not masked by other frequency components in the received voice;
Have
The gain generation unit
The background noise amount output from the background noise amount masking calculation unit and the received voice amount output from the received voice masking calculation unit are compared for each frequency band, and the background noise amount is larger than the received voice amount. The audio processing apparatus according to claim 1 or 2, wherein a gain applied to the second audio signal in a corresponding frequency band is set to 1 or more. The audio signal input from the microphone is divided into a plurality of first audio signals in a predetermined frequency band, the first audio signal for each frequency band is weighted according to a signal-to-noise ratio, and the audio signal Estimating a background noise amount for each of the divided frequency bands included in
The received voice signal is divided into a plurality of second voice signals for each predetermined frequency band, and based on the estimated background noise quantity for each frequency band, the larger the background noise quantity, the corresponding second voice signal becomes. Set a gain to be applied to the second audio signal for each frequency band so as to increase,
The received audio signal is corrected by multiplying the corresponding second audio signal of the frequency band by the gain,
A speech processing method for synthesizing the frequency of the corrected second speech signal for each frequency band and reproducing the corrected received speech signal.   The audio processing method according to claim 4, wherein a gain for each frequency band is limited to a predetermined upper limit value or less. Calculate the amount of background noise for each frequency band that can be heard by humans, not masked by other frequency components in the background noise,
Calculate the amount of received voice for each frequency band that can be heard by humans that is not masked by other frequency components in the received voice,
The background noise amount and the received voice amount are compared for each frequency band, and when the background noise amount is larger than the received voice amount, the gain applied to the second voice signal in the corresponding frequency band is set to 1 or more. The voice processing method according to claim 4 or 5, wherein the voice processing method is set.

Images