JP3394506B2 - Voice discrimination device and voice discrimination method - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voiced / unvoiced sound discriminating device (speech discriminating device) of a voice encoding / decoding device used when digitally transmitting or storing voice, and a discrimination method thereof.

[0002]

2. Description of the Related Art As a conventional voiced sound / unvoiced sound discriminating device (voice discriminating device) of this type, for example, Japanese Patent Laid-Open No. 61-2780.
In the above device, the sum of the low-order terms of the cepstrum is used as a discrimination parameter for voiced sound and unvoiced sound, and the discrimination result is a binary value of voiced sound and unvoiced sound.

FIG. 6 shows the conventional voiced sound shown in the above document.
FIG. 7 is a configuration block diagram of an unvoiced sound discrimination device (speech discrimination device), and FIG. 7 is a diagram illustrating a distribution of discrimination parameters 20 of the voiced sound / unvoiced sound discrimination device (speech discrimination device) of FIG. 6.
In the figure, 18 is a cepstrum, 19 is an addition circuit, 20 is a discrimination parameter, 21 is a threshold comparison circuit, and 22 is a discrimination result.

The operation of the voiced sound / unvoiced sound discrimination apparatus (voice discrimination apparatus) shown in FIG. 6 will be described below with reference to FIGS. First, the adder circuit 19 receives the input cepstrum 18
Then, the sum of the low-order terms of is obtained and this is output as the discrimination parameter 20. The threshold comparison circuit 21 determines that the input discrimination parameter 20 is an unvoiced sound when the discrimination parameter 20 is less than a predetermined fixed threshold, and the voiced sound when the discrimination parameter 20 is the fixed threshold or more, and outputs a discrimination result 22.

FIG. 7 shows that the power of the background noise is so large that it cannot be ignored compared with the power of the voice signal, that is, the noise level is high, and the power of the background noise is so small that it can be ignored compared with the power of the voice signal, that is, noise. It shows a model of the distribution of the discrimination parameter 20 when the level is low.
In the figure, a curve A is an unvoiced sound when the noise level is low, and a curve C
Is the unvoiced sound when the noise level is high, curve D is the distribution of curves A and C combined, and curve B is the distribution of voiced sounds. The distribution of the discrimination parameter 20 in voiced sound does not change significantly depending on the noise level. Now, if the fixed threshold for optimally separating the unvoiced sound A and the voiced sound B when the noise level is low is E1, there is an increase in the discrimination error that the unvoiced sound C is discriminated as the voiced sound B when the noise level is high. on the other hand,
When the fixed threshold for optimally separating the unvoiced sound C and the voiced sound B when the noise level is high is E2, the discrimination error for distinguishing the voiced sound B from the unvoiced sound A increases when the noise level is low. If E3 is a fixed threshold value that optimally separates the unvoiced sound D and the voiced sound B, then E1 is set to a low noise level.
It is clear that the discrimination error increases when the noise level is high compared to when E2 is used as the fixed threshold. Also,
Regardless of which of the above thresholds is used, when the discrimination parameter 20 is a value near the threshold, there are many discrimination errors and the reliability is low.

[0006]

The conventional voiced sound / unvoiced sound discriminating apparatus (speech discriminating apparatus) is configured as described above, and since the discrimination parameter is only the sum of the low-order terms of the cepstrum, the discrimination parameter is It is said that there are many discrimination errors when the value is close to the discrimination threshold, and the discrimination error increases when the speech has a noise level different from the assumed background noise level when setting the discrimination threshold for discriminating voiced sound and unvoiced sound. There are challenges.

The present invention has been made to solve the above problems, and provides a voiced / unvoiced sound discriminating apparatus (speech discriminating apparatus) and its discriminating method that do not depend on the level of the background noise level and have few discrimination errors. The purpose is to get.

[0008]

A speech discrimination apparatus according to the present invention divides an input speech frame into a plurality of subframes, and obtains a noise level for each subframe based on a comparison between input speech power and noise power. Input the noise level determination means and the noise level obtained for each subframe,
Threshold calculation means for obtaining a threshold for detecting a voice section for each subframe, and collating means for determining whether or not the input voice frame is a voice section using each threshold obtained for each subframe. It is characterized by having and.

The noise level determining means outputs a binarized noise level, and the threshold calculating means stores in advance two constants corresponding to each value of the binary determination, and the two constants are stored in advance. It is characterized in that a constant corresponding to the binarized noise level value is selected from among the above, and the threshold is calculated using the selected constant.

The noise level judging means outputs a multi-valued noise level of 3 or more, and the threshold calculating means stores in advance a plurality of constants corresponding to the respective values of the multi-value judgment, A feature is that a constant corresponding to the value of the multileveled noise level is selected from a plurality of constants and the threshold is calculated using the selected constant.

A speech discrimination method according to the present invention divides an input speech frame into a plurality of subframes, and determines a noise level for each subframe based on a comparison between input speech power and noise power, and a noise level determination step, The noise level calculated for each subframe is input, the threshold calculation step for calculating the threshold for detecting the voice section for each subframe, and the input speech frame is calculated by using each threshold calculated for each subframe. And a collation step of determining whether or not it is a voice section.

The noise level determining step outputs a binarized noise level, and the threshold calculating step stores in advance two constants corresponding to each value of the binary determination, and the two constants are stored in advance. It is characterized in that a constant corresponding to the binarized noise level value is selected from among the above, and the threshold is calculated using the selected constant.

The noise level determination step outputs a multi-valued noise level of 3 or more, and the threshold value calculation step stores a plurality of constants corresponding to the respective values of the multi-level determination in advance. A feature is that a constant corresponding to the value of the multileveled noise level is selected from a plurality of constants and the threshold is calculated using the selected constant.

[0014]

In the embodiments described below of the present invention, in a voiced sound / unvoiced sound discrimination device (speech discrimination device) for discriminating between voiced sound and unvoiced sound of a voice signal, a discriminant parameter value obtained by analyzing an input voice frame is obtained. Based on, voiced sound, unvoiced sound, select the discrimination condition for discriminating silence from a plurality of different discrimination conditions, further, according to the selected discrimination condition, power, peak value of the normalized autocorrelation, the number of zero crossings, First-order linear prediction coefficient, determination result of past speech frame,
Using at least one of the low-order terms of the cepstrum as a discriminant parameter, collating with a predetermined threshold value, the above voiced sound, unvoiced sound, by operating the collating means for outputting the discrimination result of unvoiced voiced sound, unvoiced sound, It is possible to reduce the discrimination error in the silence discrimination.

In the following embodiments of the present invention, in a voiced sound / unvoiced sound discriminating apparatus (speech discriminating apparatus) for discriminating between voiced sound and unvoiced sound of a voice signal, a discrimination parameter obtained by analyzing an input voice frame is used. , Unvoiced sound, match with the judgment conditions of silent judgment, if any one of the classifications, voiced sound, unvoiced sound, or silent is output as the judgment result, if it does not surely enter any of the classifications, voiced features If the collating means operates so as to output the discrimination result as quasi-voiced sound when it has, and as quasi-voiceless when it has a silent feature, voiced sound, unvoiced sound, or non-voiced intermediate quasi-voiced sound, quasi-voiced sound Silence can be output as the determination result.

In the following embodiments of the present invention, a voiced sound / unvoiced sound discriminating apparatus (speech discriminating apparatus) for discriminating between voiced sound and unvoiced sound of a voice signal obtains a background noise level of an input voice frame and outputs it as a noise level. The noise level determination means operates so as to select a discrimination condition for discriminating voiced sound, unvoiced sound, or silence from a plurality of different discrimination conditions according to the above noise level value, and analyzes the frame of the input voice. By comparing the discrimination parameter obtained as a result with a predetermined threshold to perform voiced sound, unvoiced sound, and silent discrimination, by the noise level value,
The thresholds for voiced sound, unvoiced sound, and silence discrimination can be changed.

In the following embodiments of the present invention, the noise level determination means determines at least one of the determination result, the power, and the peak value of the normalized autocorrelation between the input speech frame and the past speech frame. Used as
By comparing with a predetermined threshold value, the section corresponding to the voiced sound section and the silent section is determined for the input speech frame and the past speech frame, and the average power of the voiced section and the silent section is calculated, respectively, It is possible to determine and output the noise level by comparing the voiced sound average power with the silent voice average power and comparing the voiced sound average power with the silent voice average power.

In the embodiments of the present invention described below, the noise level determination means updates the average power of the frames in which the power of the input voice frame is larger than the average power of the past voice frames for each frame of the input voice. Calculated as the voiced sound average power, and calculated as the silent average power while updating the average power of the frames in which the power of the input voice frame is smaller than the average power of the past voice frames for each frame of the input voice, By comparing the voiced sound average power and the silent sound average power,
The level of noise level can be determined and output.

[0019]

EXAMPLES Example 1. FIG. 1 is a block diagram showing a first embodiment of a voiced sound / unvoiced sound discrimination device (voice discrimination device) and its discrimination method according to the present invention. In FIG. 1, 1 is power as a discriminant parameter obtained by analyzing a frame of an input speech, 2 is a peak value of normalized autocorrelation, 3 is the number of zero crossings, 4 is a first-order linear prediction coefficient, and 5 is noise. Level determination means, 6 noise level, 7 average silence power, 8 average voiced sound power, 9 threshold calculation means, 10 power determination threshold, 11 matching means, 12 determination result, 13 register, 14 register The power of the past speech frame, 15 is the normalized autocorrelation peak value of the past speech frame, 16 is the discrimination result of the past speech frame, and 17 is the low-order term of the cepstrum.

The operation of the voiced sound / unvoiced sound discriminating apparatus (voice discriminating apparatus) and its discriminating method shown in FIG. 1 will be described below with reference to the drawings. First, the noise level determination means 5 uses the normalized autocorrelation peak value 2 of the input speech frame, the past normalized autocorrelation peak value 15 stored in the register 13, and the determination result 16 of the past speech frame. On the other hand, a preset silent condition (for example, the normalized autocorrelation peak value 2 for 10 consecutive frames is equal to a predetermined threshold P1).
The average power of a section satisfying the following condition is determined from the power 1 of the input voice frame and the power 14 of the past voice frame, and the average power 7 is output. On the other hand, the average power voiced sound section determination condition of the section that satisfies the voiced sound section determination condition (for example, the peak value of the normalized autocorrelation for 5 consecutive frames is greater than or equal to the predetermined threshold P2) is defined as the input voice frame. 1 and the power 14 of the past speech frame, and output as the voiced sound average power 8. The difference between the silent average power 7 and the voiced average power 8 is a predetermined threshold D1.
If it is smaller, it is determined that the noise level is high, and the noise level 6 output from the noise level determination means 5 is "1".
On the other hand, when it is larger than the predetermined threshold value D1, it is determined that the noise level is low, and the noise level determination means 5
"0" is output as the noise level 6 which is the output of the.

Next, in the threshold value calculating means 9, when the input noise level 6 is "0", it is inputted from the noise level judging means 5 by the equation (1), and when it is "1", it is inputted from the noise level judging means 5. Averaged silent power 7 is PUV, voiced average power 8
Is set as PV, the discriminating threshold value of the power of the frame of the input speech is determined, and the power discriminating threshold value 10 output from the threshold value calculating means 9 shown in the equations (1) and (2) is compared with the collating means 11.
Send to.

[0022]

[Equation 1]

However, TH1, TH2, and TH3 are power discrimination thresholds, PUV is silent average power, and PV is voiced average power.

Next, in the matching means 11, the power 1 of the input speech frame, the peak value 2 of the normalized autocorrelation, the number of zero crossings 3, the first-order linear prediction coefficient 4, the sum 17 of the low-order terms of the cepstrum, and the noise. Noise level 6 from level determination means 5, power determination threshold value 10 from threshold value calculation means 9, register 13
The discrimination result 16 of the past voice frame from is input, and, for example, first, one of the following categories a or b to e is selected. In the case of a, that is, when the logical product of any one of Expressions (3) of the determination conditions is satisfied, it is determined as an unvoiced sound and the determination result 12 is output. In case of b to e, b to e
The threshold calculation means 9 determines which one of the categories is selected.
It is determined by the magnitude relationship between the value of TH, which is the power determination threshold value 10 from the above, and the value of POW, which is the power 1 of the input voice frame. In the above, when a can be identified as unvoiced sound,
b is a case where the probability of voiced sound is high, c is a case where the probability of voiced sound is a little high, d is a case where the probability of silence is a little high, and e is a case where the probability of silence is high.

Next, depending on which of the above categories b to e has been selected, FIG. 2, FIG. 3, FIG. 4 and FIG.
According to the determination flow of No. 1, any of voiced sound, quasi-voiced sound, quasi-silence, and silence is determined, and the determination result 12 is output. In addition,
In the categories a to e, the discrimination conditions capable of discriminating voiced sound, unvoiced sound, and silence are different from each other. Therefore, it is necessary to set the discrimination conditions individually in each classification.
This discrimination condition is experimentally determined. Here, quasi-voiced sound is defined as a case where some of the conditions for judging as voiced sound are missing, and quasi-silence is defined as a case where some of the conditions for judging as silence are missing. To do.

[0026]

[Equation 2]

B: In the case of POW> TH1, determination is made according to FIG. In the case of c: TH1 ≧ POW> TH2, the determination is made according to FIG. In the case of d: TH2 ≧ POW> TH3, the determination is made according to FIG. If e: POW ≦ TH3, it is determined according to FIG. However, the above-mentioned discriminant of the section a, b to e in FIGS.
In 4 and 5, TH1, TH2 and TH3 are power discrimination thresholds 10 (TH1>TH2> TH3), PUV is silent average power 7, PV is voiced average power 8, and POW.
Is the power 1, AC is the peak value 2 of the normalized autocorrelation, C is the sum 17 of the low-order terms of the cepstrum, CMIN is the discrimination threshold of the sum of the low-order terms of the cepstrum, Z is the number of zero crossings 3, and A1 is the first-order Linear prediction coefficient 4, NL is noise level 6, VO is past speech frame discrimination result 16, T1, T11, T12, T2, T21,
T22, T23, T24, T3, T31, T32, T33, T34, T4, T41, T42, T43, T44
All represent fixed thresholds.

Next, in the register 13, the power 1 of the input speech frame, the peak value 2 of the normalized autocorrelation, the accumulated power of the past 10 frames, the normalized peak value of the autocorrelation, and the discrimination result of the collating means are displayed. Update.

Example 2. In the first embodiment, the power discrimination threshold is determined based on the silent power and the voiced average power, but the power discrimination threshold may be determined from the maximum value of the power of the past speech frame, for example, by the formula (4). Is also possible.

[0030]

[Equation 3]

However, in the equation (4), TH1, TH
2, TH3 represents a power discrimination threshold value, and Pmax represents, for example, the maximum power value in the past 30 frames. It is also possible to correct the power discrimination threshold value obtained from the average silent power and the average voiced sound power by using the maximum value of the power in the past speech frame, or to correct the discrimination result of voiced sound, unvoiced sound, and silence. is there.

Example 3. In the first embodiment, silence determination is performed according to the peak value of the normalized autocorrelation function, the determination result of the past speech frame, and the noise level according to FIG. 2, but it is determined to be silence in the past by using the low-order term of the cepstrum coefficient, for example. It is also possible to obtain the spectrum outline of the generated frame and perform silence determination based on this spectrum outline and the distance of the spectrum of the frame of the input voice.

Example 4. In the first embodiment, the discrimination is performed using the discrimination parameter obtained by analyzing each frame of the input voice. However, the frame of the input voice is divided into a plurality of subframes, and the subframe is analyzed and obtained. It is also possible to make a determination using the parameters that are set or correct the determination result.

Example 5. In the first embodiment, the power of the frame of the input voice is used as the discrimination parameter for discriminating the discrimination condition, but it is also possible to use the sum of the low-order terms of the cepstrum.

Example 6. In the first embodiment, the noise level is set to 2
Although the values are discriminated, it is also possible to make them multi-valued or continuous numerical values.

Example 7. In the first embodiment, it is possible to include the maximum in-frame amplitude value in the discrimination parameter.

[0037]

As described above, according to the present invention, there are few discrimination errors even in the vicinity of the discrimination threshold of the discrimination parameter obtained by analyzing the frame of the input speech, and the background noise level does not depend on the level. It is possible to obtain a voiced sound / unvoiced sound discriminating apparatus (speech discriminating apparatus) and a discrimination method thereof with few discrimination errors. Further, it is possible to obtain a voiced sound / unvoiced sound discriminating apparatus (speech discriminating apparatus) and its discriminating method capable of discriminating an intermediate voice frame having both voiced and unvoiced characteristics.

[Brief description of drawings]

FIG. 1 is a configuration block diagram of a voiced sound / unvoiced sound discrimination apparatus (voice discrimination apparatus) showing a first embodiment of the present invention.

FIG. 2 is a diagram exemplifying discrimination conditions of a voiced sound / unvoiced sound discrimination device (voice discrimination device) of FIG.

FIG. 3 is a diagram exemplifying a discrimination condition of a voiced sound / unvoiced sound discrimination device (voice discrimination device) of FIG. 1;

FIG. 4 is a diagram exemplifying a discrimination condition of the voiced sound / unvoiced sound discrimination device (voice discrimination device) of FIG. 1;

5 is a diagram exemplifying the discrimination conditions of the voiced sound / unvoiced sound discrimination device (voice discrimination device) of FIG. 1;

FIG. 6 is a configuration diagram showing a conventional voiced sound / unvoiced sound discrimination device (voice discrimination device).

7 is a diagram showing a distribution of discrimination parameters of the voiced sound / unvoiced sound discrimination device (voice discrimination device) of FIG. 6;

[Explanation of symbols]

1 input speech frame power, 2 peak value of normalized autocorrelation, 3 zero-crossing number, first-order linear prediction coefficient, 5 noise level determination means, 6 noise level, 7 silent average power, 8 voiced average Power, 9 threshold calculation means, 10 power discrimination threshold, 11 collation means, 12 discrimination result, 13 register, 14 past speech frame power, 15 past speech frame normalized autocorrelation peak value, 16 past speech frame Judgment result, 17
Low-order term of cepstrum, 18 cepstrum, 19 adder circuit, 20 discrimination parameter, 21 threshold value comparison circuit,
2 discrimination result.

Continuation of the front page (56) References JP-A-4-100099 (JP, A) JP-A-5-130067 (JP, A) JP-A-5-224686 (JP, A) (58) Fields investigated (Int .Cl. ⁷ , DB name) G10L 11/02 G10L 15/04

Claims (2)

(57) [Claims] 1. A noise level determination means for dividing an input speech frame into a plurality of subframes, and for obtaining a noise level for each subframe based on a comparison between input speech power and noise power, and for each subframe. Threshold value calculating means for inputting a noise level and obtaining a threshold value for detecting a voice section for each subframe, and each threshold value obtained for each subframe and each subframe
Substituting a predetermined parameter obtained by analyzing the input voice
Using the result of comparison for each frame, and a collating means for making a decision input speech frame whether a voice section, the noise level determination means 3 or more multivalued the noise Les
And outputs the bell, and the threshold value calculation means outputs the composite value corresponding to each value of the multivalued judgment.
Store a number constant in advance and select from among the above constants.
Select and select the constant that corresponds to the multilevel noise level value.
A voice discrimination device characterized by calculating a threshold value using a selected constant . 2. A noise level determination step of dividing an input speech frame into a plurality of subframes, and obtaining a noise level for each subframe based on a comparison between the input speech power and noise power, and a noise level determination step for each subframe. Threshold value calculation process of inputting the noise level and calculating the threshold value for detecting the voice section for each subframe, and each threshold value calculated for each subframe and each subframe
Substituting a predetermined parameter obtained by analyzing the input voice
A comparison step of determining whether or not the input voice frame is in a voice section using the result of comparison for each frame , and the noise level determination step is a multilevel noise level of 3 or more.
The bell is output, and the threshold calculation step described above is performed for each value of the multi-valued judgment.
Store a number constant in advance and select from among the above constants.
Select and select the constant that corresponds to the multilevel noise level value.
A voice discrimination method characterized by calculating a threshold value using a selected constant .