JPH02272836A - Voice section detection system - Google Patents

Abstract

PURPOSE:To accurately estimate noise power even if noise power fluctuates large in the middle of communication, to update the power threshold and to reduce erroneous detection by updating the power threshold based on a specified system. CONSTITUTION:A power calculation part 12, a sound/silence decision part 13, a power threshold calculation part 14, a zerocross number calculation part 16 and a maximum zerocross interval calculation part 17 are provided. The power threshold calculation part 14 detects a noise part in a sound decision section based on the power P of the block of a sound input sampling signal, the number of zerocross times ZC, a maximum zerocross interval IZCmax and a sound/silence decision output VD by the difference of characteristics on frequencies between the sound and noise, and the power threshold is updated by power information on the detected noise part. Thus, fluctuated noise power can accurately be estimated even if noise power fluctuated large in the middle of communication, the power threshold can be updated, and errorneous detection by noise power fluctuation can be reduced.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a voice section detection method, and can be applied to, for example, a digital voice insertion system or a voice packet communication system in the digital communication field.

[Prior Art] FIG. 2 is a block diagram showing an example of the configuration of a conventional voice section detection method.

In FIG. 2, an audio input sample signal is input to a power calculation section 2 via an input terminal 1, and the power calculation section 2 divides it into blocks each having a fixed number of samples, and calculates the average power of the sample values of each block. Below Denkani, Pi (simply referred to as block power) is calculated. The power Pi of each block obtained in this way (i indicates the block number) is given to the voice/silence determination section 3 and the power threshold calculation section 4.

The sound/silence determining unit 3 compares the power Pi with a power threshold value Ti for determination inputted from the power threshold calculating unit 4, and when the power Pi is equal to or greater than the power threshold Ti, the block The block i is determined to be a sound block, and when the power Pi is smaller than the power threshold value Ti, the block i is determined to be a silent block and outputted to the next stage via the output terminal 5.

The power threshold calculation unit 4 updates the power threshold for each block. When the power Pi of block i is smaller than the current power threshold Ti by more than the fluctuation absorption offset value α, that is, when Pi<Ti−α, the power threshold T1+1 of the next block i+1 is set as Ti1l =
Calculated by Pi+α. Therefore, the power threshold becomes somewhat smaller.

In other words, the value Ti - α that is the lowest value of the fluctuation absorption offset value α from the power threshold Ti is defined as the minimum value of power up to the current time i, and the power fluctuation absorption coefficient α of background noise is added to this minimum value. Let the value be the power threshold value Ti÷1 of the new block i+l.

On the other hand, when the power Pi of block i is not smaller than the value obtained by subtracting the fluctuation absorption offset value α from the power threshold value Ti at that time, that is, when Pi≧Ti−α, the power of the next block i+1 is The previous power threshold Ti is used as the threshold Ti÷1.

[Problem to be Solved by the Invention 1] However, in the above method, for a silent block whose power is small to a certain extent, the minimum value of the block power is calculated, and an offset value is added to this to determine the next block. The power threshold for the next block is controlled adaptively, but for a sound block or a silent block close to the power threshold, the power threshold for determining the next block is changed from the previous one. Since it is the same as the threshold, that is, no adaptive control is performed, so if the noise power level increases significantly during a call, or if the noise power close to the power threshold becomes even larger, , a problem has arisen in which noise is erroneously determined to be active.

The present invention has been made in consideration of the above points, and eliminates the problem of continuously erroneously determining that noise is active when there is a large fluctuation in noise power. The present invention aims to provide a speech interval detection method that can accurately estimate the noise power after fluctuation even when the noise power fluctuates, enable adaptive setting of a power threshold that can follow fluctuations in noise power, and reduce false judgments. Do what you do.

[Means for Solving the Problems] The present invention provides a power calculating means for calculating the power of each block from an audio input signal divided into blocks of a predetermined length, and a power calculating means for calculating and outputting a power threshold value. The threshold calculation means compares the power calculated by the power calculation means with the power threshold output from the power threshold calculation means, and determines that there is a sound when the power is greater than the power threshold. voice/silence determination means for determining silence when the power is smaller than the power threshold; and power threshold calculation by calculating the number of zero crossings and maximum zero crossing interval of each block from the audio input signal. and zero-crossing information calculation means. The power threshold calculation means is characterized by the power threshold calculation processing that it executes.

The power threshold calculation process executed by the power threshold calculation means is performed as follows.

(a) If the section determined to be silent by the sound/silence determining means occurs for a predetermined first time or longer, calculate the average power of that section, set the minimum value as the first minimum value, and The value is multiplied by a constant noise power fluctuation absorption coefficient, and this multiplied value is used as a new power threshold.

(b) Once a sound is detected based on the judgment based on the current power threshold, this time is considered the first time and the above (a)
), the average power for a predetermined first time period is calculated, and this minimum value is determined as the second minimum value.

Also, whether the absolute value of the difference between the number of zero crossings of each block and the number of zero crossings of the previous block is within the first certain value has occurred consecutively for a first certain number of blocks or more, or Either the absolute value of the difference between the maximum zero-crossing interval and the maximum value of the maximum zero-crossing interval of the previous block is within a second certain value occurs continuously for a second certain number of blocks or more. The number of blocks is counted, and this count value is measured from the first time within a predetermined second time,
Only when the value is not equal to or greater than the predetermined value, the second
The minimum value of is set as the first minimum value in the process of (a) above, and the value multiplied by the fluctuation absorption coefficient is calculated as a new power threshold value.

(C) When the first minimum value is calculated and the power threshold is updated, and when the count value exceeds a predetermined value, the count value is cleared to zero and the second minimum value is set to the initial value. Reset to .

[Operation] Basically, in the present invention, the power calculating means calculates the power for each block of a predetermined length of the audio input sample signal, and the voice/silence determining means gives this power from the power threshold calculating means. The determination is made by comparing it with a power threshold value. Further, the power threshold calculation means updates the power threshold based on the power of the block, the number of zero crossings, the maximum zero crossing interval, and the voice/silence determination output. The present invention is characterized by the adaptive calculation control of the power threshold value executed by the power threshold calculation means.

The reason for adopting such adaptive calculation control will be explained below.

In the process (a) above, the minimum value is found because if the first time is long enough, even if the minimum value continues to fall, that value will be close to the true average power of the noise, and the power using the minimum value will be This is because the detection sensitivity by setting the threshold value becomes better.

However, if the noise power increases significantly during a call,
With the method (a) above, it is not possible to set a power threshold according to noise with increased average power.

Therefore, the following differences between noise and speech are used. If the audio input signal is noise, both the number of zero crossings and the maximum zero crossing interval in each block change randomly. However, in the vowel part of speech, periodicity appears in the waveform,
Therefore, the number of zero crossings and the maximum zero crossing interval in each block are both approximately constant.

Therefore, if the rate at which the number of zero crossings and the maximum zero crossing interval are almost constant during the second time period, which is longer than the average length of the audio, is equal to or greater than a certain value (block count value), then that section is the section that includes the audio. Considering this, the power threshold value is not updated in this section. On the other hand, if the rate at which the number of zero crossings and the maximum zero crossing interval are approximately constant within the second time period is less than a certain value, there is a noise portion during the second time period, and the average power of the noise portion is Estimation is made in the same manner as in (a) above, and the power threshold is updated using this as new noise power.

When the power threshold is updated while silence continues, or when the count value exceeds a predetermined value, the above (C) is set so that the above process (b) can be restarted from the beginning. It is set to the initial state by processing.

[Example] Hereinafter, an example of the present invention will be described with reference to the drawings.

Here, FIG. 1 is a block diagram showing this embodiment.

−の、几■ In FIG. 1, an audio input sample signal divided into blocks of a predetermined length is given to a power calculation section 12 via an input terminal 11, and the power calculation section 12 calculates the average of each block. The power P(+) (representing electric power: i is the number below the block number, simply referred to as the power of the block) is calculated and given to the voice/silence determining section 13. The voice/silence determination unit 13 is given a power threshold P th(i) from the power threshold calculation unit 14, and the power P(i) of each block is calculated as the power threshold Pth(i). In comparison, the sound/non-sound determining unit 13 makes a determination according to the following equation (1).

1 (sound) (P(i) >Pth(i)) VD(i
) = ... (1)
0 (silence) (P(i)≦Pth(i)) The judgment output VD(i) thus obtained is outputted to the next stage via the output terminal 15.

The audio input sample signal divided into blocks is also provided to the zero crossing number calculation unit 16.

The zero-crossing count calculation unit 16 calculates the zero-crossing count (reflects the frequency information of the audio input sample signal, and takes different counts for vowels and noise), which is the number of sign inversions of the sample signal in each block. (i) is calculated and given to the power threshold calculation section 14, and the zero-crossing interval of the sample signal in each block is calculated and given to the maximum zero-crossing interval detection section 17. The maximum zero-crossing interval detection unit 17 calculates the maximum zero-crossing interval (reflects the frequency information of the audio input sample signal, and takes different intervals for vowels and noise) IZCmax(i) for each block i. and gives it to the power threshold calculation section 14.

The power threshold calculation unit 14 adaptively calculates the power threshold pth based on the power P of the block, the number of zero crossings ZC1, the maximum zero crossing interval IZCmaX, and the voice/silence determination output VD. The signal is given to the sound/silence determining section 13.

The power threshold calculation method executed by the power threshold calculation unit 14 will be described below.

The power threshold P th(i) is updated by the following method (i) or method (ii).

(i) By determination using current power thresholds,
Every time a predetermined number N of blocks for which the sound/non-sound determination output VD indicates that there is no sound occur consecutively, the power threshold value pth is updated according to the following calculation formula.

P(i) = (ΣP(i-m/N...
・(2)-Q Pmin(i)=min (Pmin(i-1),
P(i)) −(3)Pth(i+1) =β−Pm
1n(i) - (4) These formulas are for the case where N consecutive blocks up to block i are silence judgment blocks, and the power threshold value pth(i+1) of the next block i+1 is calculated. This is the case. Further, β is a noise power fluctuation absorption coefficient which is a constant value larger than 1.

(ii) In a state in which the first method (+) is followed, if a voice determination is made for block m, that is, if P(m) > Pth(m), from this block m onward, method (+) is applied. Processing of method (:i) is performed in parallel. Note that the power threshold pth is not updated until a predetermined time period described later has elapsed after the process of method (ii) is started. The predetermined time, which will be described later, during which the process of method (11) is performed is the time during which preparatory calculations are performed to reflect the influence of background noise on the power threshold value pth.

In detail, in this second method, the following processes (i ia) and (i ic) are performed.

(iia) - Once a sound is detected, from time t1 onwards, the inter-block average power in that section is calculated for each predetermined number of blocks N using a method similar to equation (2), and the obtained inter-block average power is Using the power, the minimum value P' m1n(i) is determined in the same manner as in equation (3).

(iib) In parallel with this process, the number of zero crossings ZC(i)
The absolute value of the difference between and the number of zero crossings ZC(i-1) one block before is calculated, and the blocks for which the obtained absolute value is less than or equal to a predetermined constant value zth are determined to be the predetermined number N of blocks.
It is determined whether or not there are S Z th or more consecutive times.

In addition, the absolute value of the difference between the maximum zero crossing interval IZCmax(i) and the maximum zero crossing interval IZCmax(i-1) one block before is calculated, and this absolute value is less than or equal to a predetermined constant value 5IZth. The block is the predetermined number of blocks N5
It is determined whether or not the number is continuous for IZth or more.

When it is determined that the number of consecutive blocks changing in the number of zero crossings or the number of consecutive blocks changing in the maximum zero crossing interval is larger than the predetermined number of blocks N S Z th or N5IZth, the count value CZ of the vowel block counter is increased by 1. Increment by one.

Then, the count value CZ reaches the predetermined value c zth within the predetermined time TA, measured from the time t1 when the sound is heard.
If this is not the case, at time t2 when a predetermined time TA has elapsed since time t1, the minimum value P' m1n(i) at time t2 found in the process (iia) above is calculated from the left side of equation (3). The pH is set to 1 in (+), and equation (4) is executed to calculate the power threshold pth.

(+ic) However, if the power threshold pth is updated by the method in (i) above, and if the count value cl exceeds the predetermined value c zth at a certain point within the predetermined time TA, then (ii) Reset various quantities related to the method (replace p'm1n(i) with the initial value, count value C
2 is cleared to zero, and the count of successive occurrences of blocks where the difference value from the previous block in the number of zero crossings ZC(i) is within a certain level, and the block where the difference value from the previous block in the maximum zero crossing interval IZCmax(i) is within a certain level. (clears the count of consecutive occurrences to zero).

Next, the reason why the power threshold is calculated as described above will be explained.

The reason for finding the minimum value in method (i) above is that if the number of blocks N is long enough, even if you keep taking the minimum value, the value will be very close to the true average power of the noise. However, setting the threshold using the minimum value is good in terms of detection sensitivity.9 However, if the noise power increases significantly during a call,
In the method (1) above, it is not possible to set a power threshold according to the noise whose average power has increased, which increases malfunctions.

Therefore, by utilizing the following difference between noise and speech (particularly vowels), it is possible to set a power threshold according to the noise power within the voice determination interval.

If the audio input signal is noise, it does not have a periodic waveform like a vowel, so the number of zero crossings in a short block,
The value of the maximum zero-crossing interval that occurs within that block also changes randomly. On the other hand, in the vowel part of speech, periodicity appears in the waveform, so the number of zero crossings in a short time block is almost constant in successive blocks, and the maximum zero crossing interval is also almost constant.

Therefore, the proportion of blocks in which the number of zero crossings ZC(i) and the maximum zero-crossing interval IZCmax(i) are approximately constant within a certain period (the above-mentioned predetermined time TA) longer than the average length TO of the audio is a constant value (the above-mentioned predetermined time TA). If the count value CZth ) or more, that section is considered to be a section including audio, and the power threshold value is not updated in this section. On the contrary, TA for a certain period of time
The number of zero crossings ZC(i) and the maximum zero crossing interval IZCm
If the rate at which ax(i) is approximately constant is less than or equal to the constant value czth, it is assumed that there is a noise part in this - constant time TA, and the average power of that noise part is calculated using the method (i) above. Similarly, the minimum value of the average power of the number of blocks N is estimated, and this is used as a new noise power to update the power threshold pth.

That is, the noise portion within the voice determination interval is detected based on the difference in frequency characteristics between speech and noise, and the power threshold is updated based on the power information of the detected noise portion.

According to the above embodiment, if silence occurs in the determination based on the current power threshold, the power threshold is updated based on the power of the silent section; Once the sound has been determined by the power threshold, the block when the change in the number of zero crossings from the previous block is within a certain value for a certain period of time or more, or the The total number of blocks in which the variation of the maximum zero-crossing interval from the previous block is within a certain value for a certain period of time is used to detect periodicity due to sound presence, and this percentage of sound presence remains constant within a certain time. When the value is less than or equal to this value, it is assumed that there is a noise part within this certain period of time, the noise power of this part is estimated by the minimum value of the average power of a predetermined number of blocks, and the power threshold is updated based on this. Therefore, even if the noise power fluctuates greatly during a call, the power threshold can be updated by correctly estimating the fluctuating noise power, and erroneous detection due to noise power fluctuations can be reduced.

Incidentally, the present invention can be applied not only to various systems in the digital communication field, but also to various systems that require detection of voice sections.

In addition, the N blocks for which the inter-block power average value is calculated according to equation (2) are blocks 1 to N, N+1 to 2N, .
You may also select blocks 1 to N,
2 to N+1.3 to N+2, . . . may be selected.

[Effect of the invention 1 As described above, according to the present invention, if silence occurs in the judgment based on the current power threshold, the power threshold is updated based on the power of the silent section; Once a voice is determined by the power threshold, the noise part within the voice judgment interval is detected based on the difference in frequency characteristics between the voice and the noise, and the power of the detected noise part is determined. Since the power threshold is updated based on the information, even if the noise power fluctuates greatly during a call, the power threshold can be updated by correctly estimating the fluctuating noise power. It is possible to reduce false detection caused by

[Brief explanation of drawings]

FIG. 1 is a block diagram showing an embodiment of the voice section detection method according to the present invention, and FIG. 2 is a block diagram showing a conventional method. 12... Power calculation unit, 13... Sound/no-sound determination unit,
14... Power threshold calculation unit, 16... Zero crossing frequency calculation unit, 17... Maximum zero crossing interval calculation unit.

Claims (1)

[Scope of Claims] Power calculation means for calculating the power of each block from an audio input signal divided into blocks of a predetermined length; power threshold calculation means for calculating and outputting a power threshold; The power calculated by the power calculation means is compared with the power threshold output from the power threshold calculation means, and when the power is greater than the power threshold, it is determined that there is a sound, and the power is Sound is determined to be silent when the power is less than the threshold value/
In the voice section detection method, the voice section detection method includes a silence determination means, and a zero crossing information calculation means that calculates the number of zero crossings and the maximum zero crossing interval of each block from the voice input signal and provides the calculated value to the power threshold calculation means. (a) If the section determined to be silent by the sound/silence determining section occurs for a predetermined first time or longer, the power threshold calculation means calculates the average power of the section, and calculates the minimum value thereof. As the first minimum value, this value is multiplied by a constant noise power fluctuation absorption coefficient, and this multiplied value is used as a new power threshold. From the time when
), calculate the average power for a predetermined first time, find this minimum value as the second minimum value, and calculate the number of zero crossings of each block and the number of zero crossings of the previous block. Either the absolute value of the difference is within the first certain value for a first certain number of blocks or more, or the maximum zero-crossing interval of the block and the maximum value of the maximum zero-crossing interval of the previous block. Count the number of blocks that satisfy either of the following conditions: the absolute value of the difference is within a second constant value, which occurs continuously for a second constant number of blocks, and this count value is measured from the first time. only when the value is not equal to or greater than the predetermined value within the predetermined second time, the second minimum value obtained at the time when the second time has elapsed from the first time is set to the first value. As the minimum value, multiply this value by the fluctuation absorption coefficient and calculate it as a new power threshold, (c) When the first minimum value is calculated and the power threshold is updated, and the count value A voice section detection method characterized in that when the count value becomes equal to or greater than a predetermined value, a count value is cleared to zero and a second minimum value is reset to an initial value.