JPH11133997A - Equipment for determining presence or absence of sound - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide the high-accuracy equipment for determining the presence or absence of sound and capable of correctly determining the absence of sound at the silent zone even under conditions such as the higher noise level, or the lower S/N ratio (sound signal to noise power ratio). SOLUTION: A power enumerating device 301 enumerates power for every definite zone (frame) in an input sound signal. A silent power presuming device 302 presumes silent power. A plurality of the resulting parameters are individually used. Each individual multivalued logic determining device 304-307 individually determines the definition of the presence or absence of sound. An overall determining device 308 carries out multivalue determination in the degree of the presence of sound, using a plurality of the individual determination results. Binary encoding is applied to the presence or absence of sound according to the determination result of the presence of sound after hangover. The final determination result is then generated.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice / silence determination used for determining voice / silence of an input voice at a stage prior to an encoder in a voice coding apparatus essential for a digital mobile communication terminal such as a digital portable telephone. Related to the device.

[0002]

2. Description of the Related Art A sound / non-speech judging device judges a sound (with sound) / no sound (no sound) of an input sound signal, and judges that there is sound (with sound) when performing voice communication. The transmission efficiency can be improved by transmitting the signal only in the section where the signal has been transmitted, and the power consumption at the time of transmission can be reduced by not transmitting (not transmitting) the signal in the silent (no sound) section. There is a demand for a sound / silence determination device (eg, a mobile phone, a portable wireless device, and the like) that performs high-accuracy sound / silence (voice presence / absence) determination.

[0003] As one of the conventional sound / non-sound judging devices, '9
There is a technology described in the IEICE Spring Conference, B-373, "Study of Speech and Silence Judgment Method". Hereinafter, the conventional sound / silence determination device will be described with reference to FIG. FIG. 5 is a block diagram showing the configuration. Reference numeral 501 denotes a power calculator for calculating the audio signal power of the input audio signal for each fixed section. Reference numeral 502 denotes the audio signal power obtained by the power calculator 501 and the past. A silent power estimator for estimating power in a silent section using the value of the voiced / silent determination result, and 503 to 505 perform individual determination according to the likelihood of voiced / silent (voice presence / absence) in the input audio signal. A plurality of individual multi-valued logic decision units 506 are integrated decision units for determining sound / silence (with or without voice) by multi-valued logic based on the decision results obtained by the plurality of multi-valued logic decision units. A hangover processor 508 delays the determination from sound to silence with respect to the determination result obtained by the comprehensive determiner 506. A hangover processor 508 uses a continuous value representing the degree of sound obtained through the hangover processor 507 to generate a sound (voice is present). ) Or silence (voice ) Is a binarizing unit determines that either. The operation of the conventional sound / non-speech determination device configured as described above will be described.

In FIG. 5, a power calculator 501 calculates the power of each fixed section (hereinafter referred to as a "frame") of an input speech signal, and a silence power estimator 502 calculates the input power obtained by the power calculator 501. The estimated silence power is updated using the audio signal power and the result of the sound / silence determination in the past. The estimated silence power is obtained by updating the estimated silence power so that the greater the probability of silence of the target frame, the closer to the value of the input signal power at that time. Then, using the input voice power obtained by the power calculator 501 and the estimated silent power obtained by the silent power estimator 502, an absolute determiner 503, a change determiner 50
4, by each individual multi-valued logic decision unit of the relative decision unit 505,
The likelihood of sound / no sound is individually determined. Among the individual multi-valued logic decision units, the absolute decision unit 503 uses the absolute value of the input signal power, and the change decision unit 504 uses the amount of change in the input signal power (difference or ratio for each frame, etc.) to determine the sound level. Make a decision,
The relative determiner 505 determines the ratio or difference of the input signal power to the estimated silent power as a parameter. Then, using the plurality of individual determination results obtained by the individual multi-valued logical determinations, the integrated determiner 506 makes a multi-valued determination of the sound level (the degree of presence of voice). The hangover processor 507 performs control such that the higher the sound level is, the longer the determination from sound to no sound is made based on the result of the overall judgment unit 506. Finally, the sound determination result after the hangover is binarized into a sound (with sound) or a silence (no sound) and output as a final determination result.

[0005] The above-mentioned conventional sound / non-speech judging device is a system in which judgment is made using only the power of an input speech signal as a parameter, and there is no need to use other parameters used in speech encoding. The determination can be made independently, and the amount of calculation required for calculating the parameters used for the determination can be very small. In addition, while using only the power of the input audio signal, a good determination can be made under various ambient noise conditions.

[0006]

However, since the above-described conventional sound / silence determination device makes a determination based on the power of an input signal, the sound / noise determination under a high-level noise environment such as noise in an automobile is performed. If there is a limit to silence determination and there is a very large level of ambient noise, even in a silent section where there is no sound, it may be determined that there is sound because the noise level is high, and transmission efficiency may decrease. There was a problem that there is.

The present invention solves the above-mentioned conventional problems. Even when the noise level is high or the S / N ratio (voice signal to noise level ratio) is low, the silent section is more correctly determined. It is an object of the present invention to provide a high-precision sound / silence determination device capable of determining silence.

[0008]

SUMMARY OF THE INVENTION In order to solve the above problems, the present invention provides a power calculator for calculating an audio signal power for each frame with respect to an input audio signal, and a degree of pitch periodicity of the input audio signal. A pitch parameter calculator for calculating a pitch parameter; and a determiner for determining whether the input voice signal is sound (with voice) or silent (no voice) from the voice signal power and the pitch parameter. When changing from voiced (with voice) to silence (no voice), when the value indicated by the pitch parameter is lower than a predetermined threshold value, the frame is determined to be voiceless (no voice). It is.

Further, the present invention provides a speech coder for performing speech coding based on the result of speech / non-speech determination for each frame, and a pitch parameter calculator for calculating a pitch parameter representing a degree of pitch periodicity of an input speech signal. And a power calculator for calculating a speech signal power for each frame with respect to the input speech signal, and a pitch parameter obtained by the pitch parameter calculator in the speech encoding device, a fixed frame A delay unit that delays by several minutes, and a determiner that determines whether there is sound (with voice) or no sound (without voice) from the voice signal power and the pitch parameter that is delayed by a certain number of frames. When the input audio signal changes from a sound (with sound) to a silence (no sound), the value indicated by the pitch parameter becomes a predetermined threshold value. Ri when low, is obtained so as to determine the frame and silence (no speech).

Further, the present invention provides a power calculator for calculating an audio signal power for each frame of an input audio signal, and a silent section using the audio signal power obtained by the power calculator and the result of the previous sound / silence determination. A silence power estimator for estimating the power of the speech signal, a delay unit for outputting a pitch parameter obtained by delaying the pitch parameter obtained by the speech encoding device at the subsequent stage by one frame, a speech signal power, the estimated silence power, and one frame A plurality of individual multi-valued logic determiners for performing individual determinations according to the likelihood of a sound or silence (with or without voice) by using a part of the pitch parameter delayed by a minute; A multi-valued logic based on the decision result obtained by the logic decision unit; In addition, in the frame determined as a sound by the individual multi-valued logic determiner, the higher the value indicating the pitch periodicity degree of the pitch parameter, the greater the certainty of the sound (with sound), and the lower the value, the lower the silence (with sound). (No sound) is determined to be large, and in a silent frame, a value indicating that it is not possible to determine whether there is sound or silence (with or without sound), or asymptotically closer to that value It is.

As described above, when the noise level is high or when S /
Even under conditions such as when the N ratio is low, a silent section can be more correctly determined to be silent.

[0012]

DESCRIPTION OF THE PREFERRED EMBODIMENTS A first invention according to claims 1 and 2 of the present invention provides a power calculator for calculating an audio signal power for each frame with respect to an input audio signal, and a power calculator for calculating an input audio signal. A pitch parameter calculator for calculating a pitch parameter representing a degree of pitch periodicity; and a determiner for determining whether there is sound (with sound) or no sound (without sound) based on the audio signal power and the pitch parameter. ,
In the determiner, the frame has a sound (with sound)
When the value of the degree of pitch periodicity indicated by the pitch parameter is lower than a predetermined threshold value when the sound level changes from silent to no sound (no sound), the noise level is determined by determining that the frame is silent (no sound). Has an effect that a silent section can be correctly determined to be silent (no sound) even under conditions such as a high S / N ratio and a low S / N ratio. The ninth aspect of the present invention realizes the first aspect by a method.

Further, the second and third aspects of the present invention.
The present invention relates to a speech code comprising a speech encoder for performing speech encoding based on a speech / non-speech determination result for each frame and a pitch parameter calculator for calculating a pitch parameter representing a degree of pitch periodicity of an input speech signal. A power calculator for calculating an audio signal power for each frame with respect to an input audio signal, a delay device for delaying the pitch parameter obtained by the pitch parameter calculator by one frame, and the audio signal power And a determiner for determining whether there is sound (with sound) or no sound (without sound) from the pitch parameter delayed by one frame of the output of the delay unit. When changing from voiced (with voice) to silence (no voice), the value of the pitch periodicity indicated by the pitch parameter is a predetermined threshold. It is lower than the value, by determining the frame as silence (no speech), in conditions such as when or if the S / N ratio the noise level is high is low,
It is possible to correctly determine a certain section of silence as silence (no speech), and at the same time, the 1 calculated by the speech encoder unit.
By using the pitch parameter before the frame, high-precision determination is performed without causing a delay due to waiting for speech coding processing of the frame to be determined and without increasing the amount of calculation required for calculating the pitch parameter. It has the effect of being able to.

The third aspect of the present invention relates to the third aspect.
The invention of the present invention is a speech coder that performs speech coding based on a sound / non-speech determination result for each frame, and a pitch parameter calculator that calculates a pitch parameter indicating a degree of pitch periodicity of the input speech signal by the speech coder. In the speech coding apparatus provided with, using a power calculator that calculates the speech signal power for each frame for the input speech signal, using the speech signal power obtained in the power calculator and the result of the previous voiced and silent determination A silence power estimator for estimating the power of the silence section, a delay unit for delaying the pitch parameter obtained by the speech encoding device at the subsequent stage by one frame and outputting the same, and the speech signal power, the estimated silence power, A plurality of individual multi-valued logics that perform individual determinations according to the likelihood of a sound or silence (with or without voice) using a part of the pitch parameters A multi-level logic based on the determination results obtained by the plurality of multi-level logic determiners, and a comprehensive determiner that determines whether there is sound or no sound (with or without voice) based on the determination result. In a frame determined by the individual multi-valued logic determiner using a parameter as a sound, the higher the pitch periodicity degree indicated by the pitch parameter of the frame, the greater the certainty of the sound. The noise level is determined by determining the likelihood of the sound, and setting the value to indicate that it is not possible to determine the presence or absence of sound or silence (with or without sound) in a silent frame, or by operating asymptotically to that value. Is high, or the S / N ratio is low, a silent frame can be correctly determined to be silent (no voice), and at the same time, the pitch parameter of the previous frame can be determined. By using such a function, it is possible to perform a highly accurate determination without causing a delay caused by waiting for the voice encoding processing of the frame to be determined and without increasing the amount of calculation required for calculating the pitch parameter. Have.

According to a fourth aspect of the present invention, in addition to the configuration of the third aspect, a delay unit for delaying a speech spectrum parameter obtained by a subsequent speech encoding device by one frame is provided. A spectrum change amount calculator for calculating the change amount of the spectrum parameter from before the frame to be determined, wherein the individual multi-valued logic determiner using the pitch parameter is used for the frame determined to be sound. Is determined that the higher the value of the pitch periodicity degree indicated by the pitch parameter of the frame, the greater the likelihood of sound, and the lower the value of the pitch periodicity degree and the smaller the amount of spectrum change, the greater the likelihood of silence. By doing so, there is an effect that it is possible to make a more accurate sound / silence determination.

Hereinafter, an embodiment of the present invention will be described with reference to FIG.
This will be described with reference to FIG. (Embodiment 1) FIG. 1 is a block diagram showing a sound / non-speech judging device according to the first invention. In FIG. 1, 101 is a power calculator for calculating an audio signal power for each frame for an input audio signal, 102 is a pitch parameter calculator for calculating a pitch parameter representing the degree of pitch periodicity of the input audio signal, Reference numeral 103 denotes a determiner that determines whether there is sound or no sound based on the audio signal power and the pitch parameter. The operation of the sound / silence determination device configured as described above will be described with reference to FIG.

In FIG. 1, a power calculator 101 calculates a speech signal power of each frame of an input speech signal, and a pitch parameter calculator 102 represents a parameter (pitch parameter) representing a degree of pitch periodicity with respect to the input speech signal. Is calculated. As a parameter (pitch parameter) representing the degree of pitch periodicity, the correlation in the delay value (pitch period) that maximizes the autocorrelation of the input speech signal or the linear prediction residual signal obtained by linear prediction filtering on the input speech signal Values (normalized pitch maximum correlation value) and a prediction gain (pitch prediction gain) in a delay value (pitch cycle) that minimizes the pitch prediction error. Next, the determiner 103 determines the presence or absence of sound or silence (with or without voice) using the audio signal power and the pitch parameter obtained by the power calculator 101 and the pitch parameter calculator 102. Here, the pitch parameter is used only when detecting a change from a sound (with sound) to a silent (no sound) in a sound section. That is, it is used to determine whether the frame to be determined next to a frame determined to be voiced (with voice) is voiced (with voice) or silent (no voice), and the pitch periodicity indicated by the pitch parameter is used. When the value of the degree is lower than the threshold value, it indicates that there is no sound (no sound). A parameter representing the degree of pitch periodicity such as a normalized pitch maximum correlation coefficient and a pitch gain is a parameter representing voiced / unvoiced voice. Based on accurate detection of the rising edge of voice, continuation of voice section in voice section ( This method is suitable for detecting from speech (with speech) to silence (without speech), that is, when the pitch periodicity is high or changes to a silent section (when the pitch periodicity is continuously low). The configuration shown in the embodiment is effective. Further, the audio signal power input to the determiner is obtained by calculating the absolute value of the power of the frame to be determined, the amount of change in power from the previous frame, and the estimated silence as shown in the third embodiment. Parameters such as a ratio to the section power are calculated, and the parameters are used alone or in a combination of a plurality of them for the determination. Accordingly, the determination of voiced silence (voice presence / absence) in the voiced section is performed by using one or more parameters and pitch parameters obtained from the input voice signal power and comparing them with threshold values respectively set. , If any or all of them indicate no sound, it is determined that there is no sound (no sound). In addition,
When the normalized pitch maximum correlation value is used as the pitch parameter, the threshold value of the pitch parameter is 0.0 to
An appropriate value at 1.0 (for example, 0.4) is set. On the other hand, the presence / absence determination in the silent section uses only one or a plurality of parameters obtained from the input audio signal power, and compares one or all of them with thresholds set for each. If it indicates that the sound is sound, it is determined that there is sound (there is sound).

As described above, according to the embodiment of the present invention, a pitch parameter calculator for calculating a pitch parameter representing the degree of pitch periodicity of an input voice signal is provided, and the input voice signal is By using the pitch parameter in addition to the input audio signal power to determine when the sound changes from “with voice” to “silence” (no voice), it is possible to determine whether the noise level is high or the S / N ratio is low. Also, the silent section can be correctly determined to be silent (no sound).

(Embodiment 2) FIG. 2 is a block diagram showing a sound / non-speech judging device according to a second invention.
In FIG. 2, reference numeral 201 denotes a sound / non-speech determination device according to the present embodiment, 202 denotes a voice coder that performs voice coding based on a voice / non-speech determination result for each frame, and 203 denotes a portion provided inside the voice coder 202. A pitch parameter calculator 204 for calculating a parameter (pitch parameter) related to a pitch period necessary for voice coding; a power calculation 204 for calculating a voice signal power for each frame with respect to an input voice signal in the voiced / silence determining device 201 Device, 205 is a delay device for delaying the pitch parameter obtained by the pitch parameter calculator 203 in the voice encoder 202 by a certain frame, and 206 is the voice signal power and the output of the delay device 205 delayed by a certain frame. A determiner 207 for determining whether there is sound or no sound from the pitch parameter, and 207 delays the determination from sound to silence among the determination results obtained by the determiner 206. It is a hangover processor.

The operation of the sound / non-speech judging device configured as described above will be described with reference to FIG. In the present embodiment, a speech coder that performs speech coding based on a speech / non-speech determination result for each frame, and a pitch parameter that the speech coder calculates a pitch parameter representing a degree of pitch periodicity of an input speech signal It is assumed that the speech coding apparatus includes a calculator. The pitch period or parameters related thereto are effective for efficiently expressing a speech signal, and are essential parameters for low bit rate speech coding such as CELP (Code Excited Linear Prediction) coding. A parameter representing the degree of periodicity (pitch parameter) is calculated in the encoding process. In FIG. 2, first, a power calculator 204 calculates the power of each frame of the input audio signal. Also,
The delay unit 205 delays the pitch parameter indicating the degree of pitch periodicity calculated by the pitch parameter calculator 203 by one frame, and obtains the pitch parameter of the previous frame. As parameters representing the degree of pitch periodicity,
Minimize the correlation value (normalized pitch maximum correlation value) in the delay value (pitch cycle) that maximizes the autocorrelation of the input signal or the linear prediction residual signal obtained by the linear prediction filtering of the input signal, and the pitch prediction error Delay value
There is a prediction gain (pitch prediction gain) in (pitch cycle). Next, the deciding unit 206 decides the presence or absence of sound using the audio signal power obtained by the power calculator 204 and the delay unit 205 and the pitch parameter of the previous frame. Here, the pitch parameter is used when detecting a change from voice to silence in a voice section.
That is, it is used to determine whether the frame to be determined next to the section determined to be voiced is voiced or silent, and is silent when the pitch periodicity degree indicated by the pitch parameter is lower than a threshold value. Indicates that When the normalized pitch maximum correlation value is used as the pitch parameter, the threshold value of the pitch parameter is set to an appropriate value in the range of 0.0 to 1.0 (for example, 0.4). By limiting the use of the pitch parameter to only the determination in a certain section determined to be sound, a parameter representing the degree of the pitch period obtained in the encoding process of the pitch period related parameter of the previous frame in the speech encoder is used. Accordingly, the determination using the pitch parameter can be performed without causing a delay due to waiting for the voice coding processing of the frame to be determined and without increasing the amount of calculation required for calculating the pitch parameter. Further, the audio signal power input to the determiner is obtained by calculating the absolute value of the power of the frame to be determined, the amount of change in power from the previous frame, and the estimated silence as shown in the third embodiment. Parameters such as a ratio to the section power are calculated, and the parameters are used alone or in a combination of a plurality of them for the determination. Accordingly, the sound / non-speech determination in the sounded section is performed by using a single or a plurality of parameters obtained from the input signal power and the pitch parameter, and comparing them with thresholds set respectively.
If any or all of them indicate silence, it is determined to be silence. On the other hand, the voiced / silent determination in the section that is determined to be silent uses only one or a plurality of parameters obtained from the input signal power, and compares them with thresholds set for each of them. Is determined to be voiced when indicates that it is voiced.

As described above, according to the embodiment of the present invention, a pitch parameter representing the degree of pitch periodicity is used as one of the determination parameters, and the pitch parameter is changed from voiced to voiceless in a voiced section. Is used in addition to the input audio signal power only when detecting the change in the sound section, so that even when the noise level is high or the S / N ratio is low, the silent section can be correctly determined to be silent.
At the same time, by using a pitch parameter calculated by the speech encoder for the previous frame, there is no delay due to waiting for speech encoding processing of the frame to be determined, and an increase in the amount of calculation required for the calculation of the pitch parameter And a highly accurate determination can be made.

(Embodiment 3) FIG. 3 is a block diagram showing a sound / non-speech judging device according to a third invention.
In FIG. 3, reference numeral 301 denotes a power calculator for calculating an audio signal power for each frame with respect to an input audio signal, and 302 denotes silence using the audio signal power obtained by the power calculator 301 and the result of the previous sound / non-speech determination. A silence power estimator for estimating the power of the section, 303 is a delay unit for delaying the pitch parameter obtained by the subsequent speech encoder by one frame and outputting it as a pitch parameter of the previous frame, and 304 to 307 are probabilities of speech and silence. Among a plurality of individual multi-valued logic decision units that perform individual decision according to the audio signal power, 304 is an absolute decision unit that determines the degree of sound with the absolute value of the power using the audio signal power, 305 is using the audio signal power A change determiner 306 for determining the degree of sound based on the power change amount, a relative determiner 306 for determining the degree of sound using the audio signal power and the estimated silence power, and 307 a pitch pattern for the previous frame. A determining pitch periodicity determining unit a sound degree with meter, 30
Reference numeral 8 denotes a comprehensive determiner for determining voiced / silent (with or without voice) by multivalued logic based on the determination results obtained by the plurality of multivalued logical determiners, and reference numeral 309 denotes a result obtained by the comprehensive determiner 308. A hangover processor for delaying the determination from speech to silence with respect to the determination result, and a binarization unit 310 for determining the degree of speech in the continuous value expression obtained through the hangover processor 309 as either speech or silence. It is a vessel.

The operation of the sound / non-speech judging device configured as described above will be described with reference to FIG. In the present embodiment, a speech coder that performs speech coding based on a speech / non-speech determination result for each frame, and a pitch parameter that the speech coder calculates a pitch parameter representing a degree of pitch periodicity of an input speech signal It is assumed that the speech coding apparatus includes a calculator. The pitch period or parameters related thereto are effective for efficiently expressing a speech signal, and are essential parameters for low bit rate speech coding such as CELP (Code Excited Linear Prediction) coding. A parameter representing the degree of periodicity (pitch parameter) is calculated in the encoding process. In FIG. 3, a power calculator 301 calculates the power of each frame of the input audio signal, and a silence power estimator 302 uses the input signal power obtained by the power calculator 301 and the result of the previous sound / silence determination. Update estimated silence power. The estimated silence power is obtained by updating the estimated silence power so that the greater the probability of silence of the target frame, the closer to the value of the input signal power at that time. In addition, the delay unit 303 delays the pitch parameter obtained by the subsequent speech encoder by one frame to obtain the pitch parameter of the previous frame. As a parameter representing the degree of pitch periodicity, a delay value (pitch period) that maximizes the autocorrelation of the input signal or the linear prediction residual signal obtained by performing linear prediction filtering on the input signal.
, And a prediction gain (pitch prediction gain) at a delay value (pitch cycle) that minimizes the pitch prediction error. Using the plurality of parameters obtained as described above individually, the individual multi-valued logic decision units 304 to 307 individually determine the likelihood of sound / no sound. The determination result is represented by, for example, a continuous value of 0.0 to 1.0. Here, this value indicates that sound is more likely to be closer to 1.0, silence is more likely to be closer to 0.0, and that the value is closer to 0.5 indicates that both cannot be determined (determination is impossible). Of the individual multi-valued logic decision units 304 to 307, the absolute decision unit 304 first outputs a decision result with the value shown above using the absolute value of the input audio signal power. It is determined that the larger the sound level, the higher the sound level. The change determiner 305 makes a determination based on the amount of change in the input audio signal power (difference or ratio from one frame before, the maximum change value in the past several frames, and the like). Is determined. The relative determiner 306 determines the ratio or difference of the input audio signal power to the estimated silent power as a parameter, and determines that the larger the value is, the higher the sound level is, and the smaller the value is, the higher the silence level is. In the pitch periodicity determiner 307, the determination method during a sound interval and the determination method during a silent interval are different. In a voiced section, that is, in a frame determined to be voiced, the higher the pitch periodicity indicated by the pitch parameter, the greater the probability of voice presence, and the lower the pitch periodicity, the greater the probability of silence. Note that the pitch parameter obtained for each frame may be temporally smoothed using past values. As a result, an average stable determination result of the section can be obtained. In addition, it is possible to prevent the sound from being cut off at the end of the voice that transits from a section having a high pitch periodicity to a section having a low pitch periodicity. On the other hand, in a silent section, that is, in a section in which the preceding frame is silent, a value indicating that determination of presence / absence of speech is not possible or asymptotically close to the value. As described above, by using a determination method that does not require the input of a new pitch parameter during a silent period, it is possible to use the pitch parameter obtained in the encoding process of the pitch period related parameter of the previous frame in the speech encoder. Thus, the determination using the pitch parameter can be performed without causing a delay due to waiting for the speech encoding process of the frame to be determined and without increasing the amount of calculation required for calculating the pitch parameter. Then, after performing individual multi-valued logical judgments, the multi-value judgment of the soundness is performed by the comprehensive judgment unit 308 using the plurality of individual judgment results. This is to obtain a multi-valued (continuous value) determination result from the most probable result of voicedness and the most probable result of silence among a plurality of determination results. The hangover processor 309 performs control such that the higher the degree of sound, the longer the determination from sound to silence is delayed with respect to the result of the comprehensive determiner 308. Finally, the sound determination result after the hangover is
At 0.5 (for example, 0.5), binarization of sound or silence is performed and the result is output as a final determination result. The hangover processor 30
Instead of 9, after binarization, a process of delaying the determination from speech to silence by a certain section may be performed.

As described above, according to the embodiment of the present invention, in addition to the input speech signal power and the estimated silence power, the pitch parameter indicating the degree of pitch periodicity is obtained by the encoding process of the previous frame in the speech encoder. By performing sound / silence determination by comprehensive determination based on the plurality of multi-value determination results, a silent section can be correctly determined even when the noise level is high or the S / N ratio is low. It is possible to determine that there is no sound, and at the same time, by using the one calculated by the speech encoder in the previous frame before the decision as the pitch parameter, the delay caused by waiting for the speech encoding processing of the frame to be determined is not caused and the pitch is not changed. High-precision determination can be performed without increasing the amount of calculation required for parameter calculation.

(Embodiment 4) FIG. 4 is a block diagram showing a sound / non-speech judging device according to a fourth invention.
In FIG. 4, reference numerals 301 to 310 are the same as in the third embodiment of the third invention shown in FIG. 40
1 is a delay unit that delays a speech spectrum parameter obtained by a subsequent speech encoder by one frame, and 402 is a delay unit 401.
Is a spectrum change amount calculator that inputs the spectrum parameters delayed by one frame obtained by the above and calculates the change amount of the spectrum from the previous spectrum parameters.

The operation of the sound / non-speech judging device configured as described above will be described with reference to FIG. This embodiment relates to the invention shown in Embodiment 3 and
Except for the operation of 7, the operation is exactly the same. Pitch periodicity determiner 307
In a certain section of a sound, the higher the degree of pitch periodicity indicated by the pitch parameter, the greater the likelihood of a sound, and the lower the degree of pitch periodicity, and the smaller the amount of spectrum change, the greater the likelihood of silence. Operate to determine. Noise signals such as vehicle interior noise have a low degree of pitch periodicity, and at the same time, the signal is relatively stationary and has a small amount of spectrum change. It can be carried out.

[0027]

As described above, according to the first aspect of the present invention, a pitch parameter calculator for calculating a pitch parameter representing the degree of pitch periodicity of an input voice signal is provided. By using the pitch parameter in addition to the input audio signal power only when detecting a change from speech to silence, the silence period can be reduced even when the noise level is high or the S / N ratio is low. The effect is obtained that the sound can be correctly determined to be silent.

Further, in the second invention, a pitch parameter representing a degree of pitch periodicity obtained by a pitch parameter calculator in a speech coding apparatus is used for determination, and the pitch parameter is used in a sound section. Used when detecting a change from sound to silence, and used in addition to the input audio signal power to correctly judge a silence section as silence even when the noise level is high or the S / N ratio is low. By using the pitch parameter calculated by the speech encoder in the previous frame at the same time, there is no delay due to waiting for the speech encoding processing of the frame to be determined, and the calculation required for the calculation of the pitch parameter is also possible. The advantage is that highly accurate determination can be performed without increasing the amount.

According to the third aspect of the invention, in addition to the input speech signal power and the estimated silence power, a pitch parameter representing the degree of pitch periodicity is obtained from the encoding process of the previous frame, and a plurality of multi-value determination results are obtained. Based on the sound / silence determination based on the overall determination, a silent section can be correctly determined to be silent even under conditions such as a high noise level and a low S / N ratio. By using the one calculated in the previous frame by the speech encoder, high accuracy can be achieved without delay due to waiting for speech encoding processing of the frame to be determined and without an increase in the amount of calculation required for calculating the pitch parameter. The effect that determination can be performed is obtained.

In the fourth aspect of the present invention, the pitch periodicity judging device increases the likelihood of the sound as the degree of the pitch periodicity indicated by the parameter increases, and the degree of the pitch periodicity increases in the sounding section. By operating to determine the likelihood of silence higher as the spectrum change amount is lower and the spectrum change amount is smaller, the characteristic that the degree of pitch periodicity such as in-vehicle noise is low and the signal is relatively stationary is used, and the spectrum change amount is used. Is used for the determination of the soundness based on the pitch periodicity, so that an effect that a more accurate determination can be performed is obtained.

[Brief description of the drawings]

FIG. 1 is a block diagram of a sound / silence determination device according to a first embodiment of the present invention.

FIG. 2 is a block diagram of a sound / silence determination device according to Embodiment 2 of the present invention;

FIG. 3 is a block diagram of a sound / silence determination device according to a third embodiment of the present invention;

FIG. 4 is a block diagram of a sound / silence determination device according to a fourth embodiment of the present invention.

FIG. 5 is a block diagram of a conventional sound / silence determination device.

[Explanation of symbols]

 101 Power Calculator 102 Pitch Parameter Calculator 103 Judge 201 Speech / No-Speech Judgment 202 Speech Encoder 203 Pitch Parameter Calculator 204 Power Calculator 205 Delayer 206 Judge 207 Hangover Processor 301 Power Calculator 302 Silence Power Estimator 303 Delay unit 304 Absolute judgment unit 305 Change judgment unit 306 Relative judgment unit 307 Pitch periodicity judgment unit 308 Total judgment unit 309 Hangover processor 310 Binarization unit 401 Delay unit 402 Spectral change calculation unit

Claims (10)

[Claims] 1. A power calculator for calculating a voice signal power for each fixed section of an input voice signal, a pitch parameter calculator for calculating a pitch parameter for each fixed section, and an output of the voice signal power and the pitch parameter. A sound / silence determination device comprising a judgment unit for judging a sound / no-sound state for each of the predetermined sections based on the above. 2. The method according to claim 1, wherein when determining whether a sound is present or not in a next second fixed section following the first fixed section of sound, the determination section corresponds to the second fixed section. A sound / non-speech determining apparatus characterized in that when the output value of the pitch parameter calculator is lower than a predetermined threshold value, the second fixed section is determined as a silent fixed section. 3. A speech encoder having a pitch parameter calculator for calculating a pitch parameter for each fixed section of an input speech signal, a power calculator for calculating a speech signal power for each fixed section of the input speech signal, A delay unit that delays the output of the pitch parameter calculator for a predetermined time; and a determiner that determines a sound or silence state for each of the predetermined sections based on an audio signal power obtained by the power calculator and an output of the delay unit. Sound / silence determination device provided with 4. The method according to claim 3, wherein the determining unit determines whether a sound is present or not in a next second fixed section following the first fixed section, and corresponds to the first fixed section. A sound / silence determining device for determining the second constant interval as a silent constant interval when a value of a delay device output obtained by delaying an output value of the pitch parameter calculator is lower than a predetermined threshold value; . 5. A speech encoder having a pitch parameter calculator for calculating a pitch parameter of an input speech signal for each fixed section, and a power calculator for calculating a speech signal power of the input speech signal for each fixed section. A silent power estimator for estimating the power of a silent section using the voice signal power obtained by the power calculator and the result of the previous sound / silence determination, and a pitch parameter obtained by the pitch parameter calculator by one frame delay. A delay unit that outputs the input voice signal and a part of the calculated voice signal power, the estimated voiceless power, and the pitch parameter delayed by one frame. A plurality of individual multi-value logic decision units for performing individual multi-value logic decision, and a multi-value theory based on a decision result obtained by the plurality of individual multi-value logic decision units The input audio signal is configured by a voiced / silent determining unit including a comprehensive determining unit that determines whether the voice is voiced or unvoiced, and the individual multi-valued logic determining unit uses a pitch parameter of the delay unit output. And having a pitch periodicity determiner to determine the pitch periodicity,
A sound / non-speech judging device for judging by individual multi-valued logic. 6. The sound / non-speech determining unit according to claim 5, wherein the individual multi-valued logic determiner using a pitch parameter in the input voice signal determines a pitch in a certain section determined as a voice of the input voice signal. The higher the value indicating the degree of pitch periodicity indicated by the parameter, the greater the certainty of the sound,
The lower the lower, the greater the likelihood of silence is determined, and in a certain interval of silence, a value indicating that determination of existence / non-speech is impossible is performed, or the operation is performed so as to approach the value. Sound / silence determination device. 7. A speech / silence determination device according to claim 5, wherein a speech encoder for outputting a speech spectrum parameter of an input speech signal and a delay unit for delaying and outputting the speech spectrum parameter output by one frame. And a spectrum change amount calculator for calculating a change amount of a voice spectrum parameter from the delay unit, and an individual multi-valued logic determiner using a pitch parameter determines in a sound section,
The higher the value of the pitch parameter is, the larger the probability of sound is determined, and the lower the value of the pitch parameter and the smaller the amount of spectrum change, the greater the probability of silence is determined. Sound / silence determination device. 8. The sound / silence determination device according to claim 1, wherein a normalized pitch maximum correlation value or a pitch prediction gain is used as a pitch parameter indicating a degree of pitch periodicity of the input audio signal. 9. A voice signal power and a pitch parameter for each fixed section of the input voice signal are calculated, and based on the calculated voice signal power and the pitch parameter, a sound or silence state for each of the predetermined sections is determined, Particularly, in the determination, when the value of the pitch parameter is lower than a predetermined threshold value when the input voice signal changes from voiced to voiceless, the predetermined interval is determined to be voiceless and voiceless. Judgment method. 10. A recording medium such as a magnetic disk, a magneto-optical disk, a ROM cartridge, or the like in which a program that implements the sound / non-sound determining device according to claim 1 by software is recorded, or using the recording medium. A device that operates as a sound / silence determination device.