JPH0795238B2 - Voiced determination device and voiced determination method - Google Patents

Abstract

Apparatus for detecting a fundamental frequency in speech utilizing a plurality of voiced detectors and selecting (104, 105) one of those detectors to make the voicing decision utilizing distance measurement values with each value generated by one of the voiced detectors. The voiced detector selected is the one which generated the best distance measurement value. The distance measurement value may be the Mahalanobis distance value or Hotelling's two-sample T<2> statistic. Two types of voiced detectors are disclosed: statistical voiced detectors (103) and discriminant voiced detectors (107). The disclosed statistical voiced detector adapts to changing speech environments by detecting changes in the voice environment in response to classifiers that define certain attributes of the speech.

Description

Description: TECHNICAL FIELD The present invention relates to determining whether speech has the presence of a fundamental frequency. This is also called voice discrimination. More particularly, the present invention relates to a selection of voiced sound detectors for simultaneously processing speech samples to make speech discrimination,
Relates to selections made based on distance measure calculations.

BACKGROUND AND PROBLEMS In low bit rate voice coders, voice quality degradation often occurs due to inaccurate voice discrimination. The difficulty in making these speech discriminations accurately lies in the fact that a single speech parameter, a classifier, is not reliable in distinguishing between voiced and unvoiced speech. The use of multiple voiced sound detectors and the selection of one of these detectors for determining whether the voice is voiced or unvoiced is described by CPCambell et al. Paper "Application of voiced / unvoiced speech to US Government LPC-10E algorithm" IEEE Acoustics
International Conference on Speech and Signal Processing, 1986, Tokyo, No. 9.
Volume 11.4, pp. 473-476 ("Voiced / Unvoiced Classific
ation of Speech with Application to the USGovern
ment LPC-10E Algorithm ", IEEE International Confer
ence on Acoustics, Speech and Signal Processing, 19
86, Tokyo, Vol.9.11.4, pp.473-476). This paper discloses the use of a multi-linear discriminant voiced sound detector that utilizes different weights and thresholds to process the same speech classifier for each speech frame. The weights and thresholds for each detector are determined by using the training data. Different levels of white noise are added to the training data for each detector. During actual speech processing, the SN ratio (Signal-to-Noise Ratio, SNR) is examined to determine the detector for speech discrimination. The range of possible values that the SNR can have is divided into sub-ranges, each assigned to one of the detectors. For each frame, the signal-to-noise ratio is calculated, the subrange is determined, and the detector attached to this subrange is selected for speech discrimination.

The problem with this method is that it does not perform well in speech environments where the speech features themselves are altered. Furthermore, the method used by Campbell is only applicable to white noise and not adjustable to color noise. Therefore, there is a need for a selection method among multiple voiced sound detectors that enables detection in a changing voice environment.

[Solution] The above problem can be solved by a voiced sound detection device selected from a plurality of voiced sound detectors by comparing separation values or merit values generated by each voiced sound detector. Progress is achieved. Separation values are also called distance measures.

The device includes different types of voiced detectors, such as a discriminant detector or a statistical detector, each of which produces a separate value.
The comparator in this device selects the voiced sound detector producing the maximum separation value in order to determine whether the voice is voiced or unvoiced. The separation value is preferably a statistical generalized distance.

All voiced sound detectors indicate whether the frame is voiced or unvoiced, and each detector first determines the discriminant variable for each of the current and previous frames. After determining the variables, each detector determines an average value for both voiced and unvoiced ones of the previous and current frames. Each detector determines the variance value for the previous and current frame voiced and unvoiced ones. After calculating the mean and variance, each detector determines a separation value from the mean and variance values for voiced frames and the unvoiced frames.

The determination of the separation values is preferably made by combining the variance values in each detector in a weighted sum. The average value of each unvoiced frame is subtracted from the average value of each voiced frame. This subtracted value is squared for each frame and the weighted sum of the variances is divided by the determined squared subtracted value. Before forming the weighted sum, each detector preferably multiplies the variance for voiced frames by the probability of occurrence of voiced frames and the variance for unvoiced frames by the probability of occurrence of unvoiced frames. Further, the squared subtraction value is multiplied by the probability of voiced frame occurrence and the probability of unvoiced frame occurrence before dividing the squared subtraction value by the weighted sum.

The method comprises calculating a first merit value defining a separation between voiced and unvoiced frames by a discriminative detector, and second defining a separation between voiced and unvoiced frames by the statistical voiced detector. Calculating a merit value of, and selecting the detector for which the best merit value has been calculated to indicate whether the frame is voiced or unvoiced.

BRIEF DESCRIPTION OF THE DRAWINGS The present invention will be readily understood by reading the following detailed description with reference to the drawings. Here: Fig. 1 is a block diagram of an apparatus using the present invention; Fig. 2 is a diagram showing the present invention in the form of a block diagram; Figs. 3 and 4 are statistical voiced voice detections of Fig. 2. Bowl 103
FIG. 5 is a more detailed view of the functions performed by block 340 of FIG. 4;

Detailed Description FIG. 1 shows an apparatus for performing an unvoiced / voiced sound discrimination operation by selecting one between two voiced sound detectors. It will be apparent to those skilled in the art to use more than two voiced sound detectors in FIG. Detector 102
The selection between and 103 is based on the distance measure generated by each detector and transmitted to the distance comparator 104. Each distance measure generated represents a merit value which indicates the accuracy of the speech discrimination of the generation detector. The distance comparator 104 compares the two distance measure values and controls the multiplexer 105 so that the detector that produces the maximum distance measure value is selected to make the unvoiced / voiced sound discrimination. However, for other scale types, the minimum merit value will dictate the detector to make the most accurate speech discrimination. The distance measure is preferably Mahalanobis distance. Detector
Preferably 102 is a discriminating detector and detector 103 is a statistical detector. However, it will be apparent to those skilled in the art that the detectors may all be of the same type and there may be more than two detectors in the system.

Consider now the overall operation of the apparatus shown in FIG. The classifier generator 101 is offset in response to each speech frame by the log of the speech energy, the log of the LPC (linear predictive analysis) gain, the log area ratio of the first reflection coefficient, and one pitch period. A classifier (cl that is preferably the square correlation coefficient of two speech segments of one frame length)
assifier) is generated. Calculations of these classifiers include digitally sampling analog speech, forming frames of digital samples, and processing these frames, which are well known to those skilled in the art. The generator 101 transmits the classifier to the detectors 102 and 103 via the passage 106.

Detectors 102 and 103 make unvoiced / voiced sound discriminations in response to classifiers received via passages 106 and multiplexes these discriminations via passages 107 and 110, respectively.
Communicate to. In addition, these detectors determine the distance measure between voiced and unvoiced frames and transmit these distances to comparator 104 via paths 108 and 109.
These distances are preferably Maharanobis distances or other generalized distances. Comparator 104 controls multiplexer 105 in response to the distance received via paths 108 and 109 so that the latter multiplexer selects the detector output producing the maximum distance.

FIG. 2 shows the statistical voiced sound detector 103 in more detail.
For each speech frame, a set of classifiers, also referred to as classifier vectors, is received from the classifier generator 101 via passage 106. Silence detector 201 is responsive to these classifiers to determine if there is speech in this frame. If speech is present, the detector 201 will transmit a signal via the path 201. If no speech is present in the frame (silence), then only then is the subtractor 207 and U / V (unvoiced / voiced) discriminator 205 activated for that particular frame. The discriminator 20 determines whether or not voice is present.
5, unvoiced sound / voiced sound discrimination is performed for each frame.

The classifier averager 202 responds to the signal from the detector 201 by averaging the classifiers for the previous frame within the classifier for the current frame to determine the individual classifiers received via the passageway 106. Maintain average. If speech (not silence) is present in the frame, silence detector 201 signals via path 210 to statistical calculator 203, generator 206, and averager 202.

Statistical calculator 203 calculates the statistical distribution for voiced and unvoiced frames. In particular, calculator 203 is responsive to the signal received via path 210 to calculate the overall probability that a frame is unvoiced and the probability that a frame is voiced. Further statistical calculator 203
Computes the statistics that each classifier would have if the frame was unvoiced and the statistics that each classifier would have if the frame was voiced. Furthermore, the calculator 203 calculates the covariance matrix of the classifier. This statistical value is preferably an average value.
The calculations performed by calculator 203 are based not only on the current frame, but also on previous frames. Statistical calculator 203 not only bases these calculations on the averages of the classifiers for the current frame received over path 106 and the classifiers received over path 211, as well as the weighting for each classifier. Based on the threshold received from the weight calculator 204 via passage 213 in determining whether the frame is unvoiced or voiced.

The weight calculator 204 includes a path 212 generated by the calculator 203.
The probability of a classifier for the current frame received via
Re-compute the weight vector a for each classifier and the threshold b for the current frame in response to the covariance matrix and the statistics. These new a and b values are then transmitted back to the statistical calculator 203 via path 213.

Weight calculator 204 also transmits weights and statistics for classifiers in both unvoiced and voiced regions to discriminator 205 via path 214 and to generator 20 via path 208.
Transmit to 6. The latter generator is responsive to this information to calculate a distance measure which is then transmitted to the comparator 104 via path 109 as shown in FIG.

U / V (unvoiced / voiced sound) discriminator 205 discriminates whether this frame is unvoiced or voiced in response to the information transmitted via passages 214 and 215, and this discriminator 11
It is transmitted to the multiplexer 105 of FIG. 1 via 0.

Shown in FIG. 2 herein, the operation of each block given in vector and matrix mathematics will be described in more detail. The averager 202, the statistical calculator 203, and the weight calculator 204 are referred to as “Estimating the Component” by NEDey.
s of a Mixture of Normal Distribution ", Biometrika, Vol. 56, No. 3, pp. 463-474, 1969). Average (decaying
Using the concept of (average), the classifier averager 202 calculates the average of the classifier for the current frame and the previous frames by calculating the following equations 1, 2 and 3:

n = n + 1 ifn <2000 (1) z = 1 / n (2) X n = (1-z) X n-1 + zx n (3) x n is a vector indicating a classification element for the current frame,
n is the number of processing frames up to 2000. z indicates the average coefficient of the breaks, X _n indicates the average of all classifiers of the current frame and the past frame. Statistical calculator 203, in response to receipt of z, x _n and X _n information, computes covariance matrix T by first computing a square and sum of product matrix Q _n as follows.

_{Q n = (1-z)} Q n-1 + zx n x 'n. (4) When Q _n is calculated, T is calculated as follows.

_{T = Q n -X n X '} n. (5) The average value is subtracted from the classifier as follows.

x _n = x _n −X _n (6) Next, the calculator 203 determines the probability that the frame represented by the current vector x _n is unvoiced by solving equation (7) as shown below. And the component of vector a is 0.391860, which corresponds to the logarithm of speech energy.
The component corresponding to the logarithm of LPC gain is −0.0520902.
And the component corresponding to the log area ratio of the first reflection coefficient is 0.5637.
The component corresponding to 082 and the square correlation coefficient is 1.361249
Preferably, b is initially equal to-8.36454.

After solving equation (7), the calculator 203 determines the probability that the classifier represents a voiced frame by solving the following equation.

P (v | _xn ) = 1-P (u | _xn ) (8) Next, the calculator 203 determines the total probability that a frame will be unvoiced by solving the equation (9) for obtaining _pn. To do.

_pn = (1-z) _pn-1 + zP (u | _xn ) (9) After determining the probability that the frame will be unvoiced,
The calculator 203 then determines two vectors u and v which give the average value of each classifier for both unvoiced and voiced frames. Vectors u and v are the statistical averages for unvoiced and voiced frames, respectively. Vector u which is the statistical average unvoiced vector
Contains the mean value of each classifier if the frame is unvoiced; and the vector v, which is the statistical average voiced sound vector, gives the mean value for each classifier if the frame is voiced. As shown below, the vector u for the current frame is solved by calculating equation (10) and the vector v for the current frame is determined by calculating equation (11).

u _n = (1-z) u _n-1 + zx _n P (u | x _n ) / _pn −zx _n (10) v _n = (1-z) v _n-1 + zx _n P (v | x _n ) / (1-p _n ) −zx _n
(11) Here, the calculator 203 transmits the vectors u and v, the matrix T, and the probability p to the weight calculator 204 via the path 212.

Weight calculator 204 is responsive to this information to calculate new values for vector a and scalar b. These new values are then transmitted back to the statistical calculator 203 via path 213. This allows the detector 103 to quickly adapt to changing environments. Even if the new values for the vector a and the scalar b are not transmitted back to the statistical calculator 203, since the vectors v and u have the latest values, the detector 103
Will continue to adapt to the changing environment. As is apparent, the discriminator 205 makes speech discrimination using the vectors u and v, and the vector a and the scalar b. When n is preferably greater than 99, vector a and scalar b are calculated as: The vector a is determined by solving the following equation.

The scalar b is determined by solving the following equation.

After calculating equations (12) and (13), the weight calculator 204
Transmits the vectors a, u, and v to the u / v discriminator 205 via the path 214. If the frame contains silence, only equation (6) is calculated.

The discriminator 205 determines whether the current frame is voiced or unvoiced in response to the transmitted information. If the component of the vector (v _n −u _n ) corresponding to the output is positive, then the frame is declared to be voiced if the following is true:

ax _n −a (u _n + v _n ) / 2> 0 (14) If the component of the vector (v _n −u _n ) corresponding to the output is negative, then the following expression is true. Then the frame is declared to be voiced.

ax _n −a (u _n + v _n ) / 2 <0 (15) Equation (14) can also be rewritten as the following equation.

_{a'x n + b-log [(} 1-p n) / p n]> 0 formula (15) can also be rewritten to the following equation.

If _{a'x n + b-log [(} 1-p n) / p n] <0 If the above conditions are not met, discriminator 205,
Declare the frame to be unvoiced. Expressions (14) and (15) represent a discrimination area for performing voice discrimination. The rewritten log terms in Eqs. (14) and (15) can be omitted if the performance is changed slightly. In this embodiment, the component corresponding to the output is preferably the log of the sound energy.

The generator 206 is responsive to the information received from the calculator 204 via the path 214 to calculate the distance measure A as follows. First, the discriminant variable d is calculated by the following equation (16).

d = a'x _n + b-log in _{[(1-p n) /} p n] (16) for generating the similar values to d for use in the next Shoshiki, various types of speech detector It is preferred to use and this will be clear to the skilled person. One such detector would be an autocorrelation detector. If the frame is voiced, then equations (17) through (20) are solved as follows.

m ₁ = (1-z) m ₁ + zd, (17) s ₁ = (1-z) s ₁ + zd ^2, (18) k ₁ = s ₁ − ▲ m ² ₁ ▼ (19) where m ₁ is It is the average for voiced frames and k ₁ is the variance for voiced frames.

The probability P _d that the discriminator 205 will declare the frame to be unvoiced is calculated by:

P _d = (1-z) P _d . (20) P _d is preferably initially set to 0.5.

If the frame is unvoiced, equations (21) through (24) are solved as follows.

_{m 0 = (1-z)} m 0 + zd (21) s 0 = (1-z) s 0 + zd 2, (22) k 0 = s 0 - ▲ m 2 0 ▼ · (23) frames is unvoiced The probability P _d that the discriminator 205 will declare is calculated by the following equation.

_Pd = (1-z) _Pd + z. (24) After calculating the equations (16) to (22), the distance measure, that is, the merit value is calculated as follows.

Equation (25) calculates the distance measure using the Hotelling 2-sample T ² statistic. For equation (25), the greater the merit value, the greater the separation. However, other merit values exist where the smaller the merit value, the greater the separation. Preferably, the distance measure may be the Mahalanobis distance given by

Preferably the third method is given by:

A fourth method for calculating the distance measure is preferably given by:

A ² = a ′ (v _n −u _n ) (28) The identification detector 102 transmits the unvoiced sound by transmitting the information indicating the voiced frame to the multiplexer 105 via the path 107 if a′X + b> 0. / Perform voiced sound discrimination. If this condition is not true then the detector 10
2 indicates unvoiced sound frame. The values for vector a and scalar b used by detector 102 are preferably the same as the initial values of a and b for statistical voiced sound detector 103.

Detector 102 determines the distance measure in a manner similar to generator 206 by performing calculations similar to those given in equations (16)-(28).

3 and 4 show the statistical voiced sound detector 103 of FIG.
The operations performed by are illustrated in more detail in flow chart form. Blocks 302 and 300 are the blocks of FIG. 2, respectively
Perform 202 and 201. Blocks 304-318 implement the statistical calculator 203. Blocks 320 and 322 implement weight calculator 204 and blocks 326-338 implement block 205 of FIG. The generator 206 in FIG. 2 is a block
Performed by the 340. Subtractor 207 is implemented by block 308 or block 324.

Block 302 computes a vector indicating the average classifier for the current frame and all previous frames. Block 300 determines if there is speech or silence in the current frame. And if there is silence in the current frame, block 324 subtracts the average for each classifier from each classifier before control is passed to decision block 326. However, if speech is present in the current frame, then blocks 304-322 perform statistical and weight calculations.
First, in block 302, the average vector is determined. Second, at block 304, a matrix of sums of squares and products is calculated. Then block 306
In, the latter matrix is used with a vector X that indicates the mean of the classifiers for the current and past frames to compute the covariance matrix T. Next, at block 308, X is subtracted from the classifier vector x _n .

Next, block 310, the current weight vector a, by utilizing the Ruibetsuko vector x _n for the current thresholds b and the current frame, the current frame is to calculate the probability of unvoiced. After calculating the probability that the current frame is unvoiced, block 312 calculates the probability that the current frame is voiced. Next, block 314 calculates the overall probability p _n that a frame will be unvoiced.

Blocks 316 and 318 compute two vectors u and v. The value contained in the vector u represents the statistical mean value that each classifier would have if the frame were unvoiced. The vector v, on the other hand, contains a value that represents the statistical mean that each classifier would have if the frame were voiced. The actual values of the classifier for the current frame and previous frames are clustered around vector u or vector v. If these frames are found to be unvoiced, the vectors representing the classifiers for the previous frame and the current frame are clustered around the vector u; otherwise the previous classifier vector is rotated around the vector v. Clustered in.

After performing blocks 316 and 318, control passes to decision block 320. If N is greater than 99 then control is passed to block 322; otherwise control is passed to block 326. Upon receiving control, block 322 then calculates a new weight vector a and a new threshold b. Vector a and value b are used by the block preceding it in FIG. 3 in the subsequent frame. Preferably, if N is required to be greater than infinity, vector a and scalar b are never changed, and detector 103 sets vectors v and u as shown in blocks 326-338. Only would respond and adapt.

Blocks 326 to 338 implement the u / v discriminator 205 of FIG. Block 326 determines whether the power term of vector v in the current frame is greater than or equal to the power term of vector u. If this condition is true, then decision block 328 is executed. The latter discriminating block discriminates whether voiced or unvoiced by a test. If it is determined in block 328 that the frame is voiced, then the frame is displayed as voiced by block 330, otherwise the frame is blocked.
Displayed as unvoiced by 332. If the power term of vector v is less than the output term of vector u for the current frame, the functions of blocks 334-338 are performed and function similarly. Finally, block 340 calculates the distance measure.

FIG. 5 illustrates, in flowchart form, the operations performed by block 340 of FIG. 4 in further detail. The determination block 501 is
The result of blocks 330, 332, 336 or 338 is examined to determine if the frame was designated as unvoiced or voiced. If the frame is designated as voiced, passage 507 is selected. Block 510 has probability P _d
And block 502 averages m for voiced frames.
₁ recalculates the, and block 503 recalculates the variance k ₁ for voiced frames. If the frame is determined to be unvoiced, decision block 501 selects path 508. Block 509 recalculates the probability P _d , block 504 recalculates the mean m ₀ for unvoiced frames, and block 505 recalculates the variance k ₀ for unvoiced frames.
Finally, block 506 calculates the distance measure by performing the indicated calculations.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP 61-48898 (JP, A) JP 60-200300 (JP, A) JP 60-114900 (JP, A) IEEE Transactions on Acoustics, Speed, and Signal Procedures Vol. ASSP-24, No. 3, June 1976, p. 201-212

Claims (23)

[Claims] 1. A voiced decision device having a plurality of detection means for executing and displaying a voiced decision of a non-training set speech frame which is either voiced, unvoiced or unvoiced, wherein each of the detection means is present or past. The apparatus has a calculation unit that calculates a merit value that defines the separation between the voiced discrimination region and the unvoiced discrimination region of the frame, and the apparatus has a detection unit that calculates the best merit value (hereinafter referred to as “best detection unit”). ) Is selected, and a voiced discrimination device for displaying the voiced discrimination of the current frame is further provided. 2. Apparatus according to claim 1, characterized in that the calculating means of each detecting means perform a statistical calculation to determine the merit value. 3. The apparatus of claim 2, wherein the statistical calculation is a distance measure calculation. 4. The calculation means determines means for identifying variables of current and past frames, means for determining an average value of current and past voiced frames, and variance values of current and past voiced frames. A means for determining, a means for determining the average value of the current and past unvoiced frames, a means for determining the variance value of the current and past unvoiced frames, and a determined average and variance value of the voiced frames and the determined 3. The apparatus according to claim 2, further comprising merit value determining means for determining a merit value from an average value and a variance value of unvoiced frames. 5. The merit value determining means, means for obtaining a sum of the variance values, weighted sum calculating means for calculating a weighted sum of the variance values, and an average value of the voiced frames for the unvoiced frames. It is characterized by comprising means for subtracting the average value, means for squaring the subtracted value, and division means for dividing the weighted sum by the squared value and generating the merit value from the division result. The device according to claim 4. 6. The weighted sum calculation means calculates a first probability that the best detection means indicates that the current frame is a voiced frame, and the best detection means causes the current frame to be an unvoiced frame. Means for calculating a second probability of indicating that the first and the variance values of the current and past voiced frames are
6. The apparatus of claim 5, comprising means for multiplying a probability and multiplying the variances of the current and past unvoiced frames by the second probability, and means for forming the weighted sum from the result of this multiplication. . 7. The apparatus according to claim 6, wherein said division means has means for multiplying said division result by said first and second probabilities to generate said merit value. 8. A statistical parameter set calculation means for calculating a set of statistical parameters in response to a set of classifiers defining speech attributes of the current frame, and the set of calculated parameters. Weight set calculation means for calculating weights corresponding to each of the classifiers in response to, and performing voiced discrimination of the current frame in response to the calculated set of weights and classifiers and the set of parameters. 8. The apparatus of claim 7, further comprising means. 9. The weight set calculation means calculates a threshold value in response to the set of parameters, and means for notifying the statistical parameter set calculation means of the set of weights and the threshold value. 9. The statistical parameter set calculation means is also responsive to a set of weights reported and a set of classifiers defining speech attributes of a frame different from the current frame. Equipment. 10. For a non-training set speech frame that is either voiced, unvoiced or unvoiced, a first representation of the degree of separation between the voiced and unvoiced discriminant regions for the current and past frames. Between a voiced discrimination region and a voiced discrimination region for the current and past frames, the first means for calculating a generalized distance value of the first frame and generating a first signal indicating that the current frame is a voiced frame. Second means for calculating a second generalized distance value representative of the degree of separation of the first frame and generating a second signal indicating that the current frame is a voiced frame; Means for selecting the first signal when it is better than the second generalized distance value, and selecting the second signal when the second generalized distance value is better than the first generalized distance value. Featuring Voiced determination device that. 11. The apparatus of claim 10, wherein the generalized distance value is a Mahalanobis distance value. 12. The means for calculating a set of statistical parameters in response to a set of classifiers defining speech attributes of a frame, and the classifier in response to the set of calculated parameters. Further comprising means for calculating weights corresponding to each of the, and means for determining the current frame to be a voiced frame in response to the calculated set of weights and classifiers and the set of parameters. The apparatus of claim 11, wherein: 13. The first means comprises means for determining the first generalized distance value in response to the set of calculated parameters and the set of calculated weights. 12 devices. 14. The apparatus of claim 13 wherein said second means is a discriminant voiced sound detector. 15. The second means determines a mean value of current and past voiced frames, a means to determine a variance value of current and past voiced frames, and a mean value of current and past unvoiced frames. Determining the variance value of the current and past unvoiced frames, and the second generalization from the determined average value and variance value of the voiced frame and the determined average value and variance value of the unvoiced frame. 15. The apparatus of claim 14, further comprising second generalized distance value determining means for determining a distance value. 16. The second generalized distance value determining means, means for calculating a weighted sum of the variance values, means for subtracting the average value of the unvoiced frames from the average value of the voiced frames, 16. The method according to claim 15, further comprising: means for squaring an averaged value and means for dividing the weighted sum of the variance values by the squared value to generate the second generalized distance value. apparatus. 17. A voiced discrimination method using first and second voiced sound detectors for executing and displaying voiced discrimination of an untrained set speech frame, comprising a voiced discrimination area and an unvoiced discrimination area of a current frame and a past frame. The first merit value defining the separation between
The step of calculating with a voiced sound detector and a second merit value defining the separation between voiced and unvoiced discriminant regions of the current and past frames
Calculating with a voiced sound detector, selecting the first voiced sound detector to display a voiced discrimination when the first merit value is better than the second merit value, and the second merit value is the second merit value. A voiced discrimination method comprising the step of selecting a second voiced sound detector to display the voiced discrimination when it is better than one merit value. 18. The step of calculating the first and second merit values comprises the step of performing a statistical calculation to determine the first and second merit values, respectively. The method of paragraph 17. 19. The method of claim 18, wherein the statistical calculation is a distance measure calculation. 20. The step of calculating the first merit value comprises the steps of determining an identification variable for each of the current and past frames, determining an average value of the current and past voiced frames, and presenting the current and past frames. Determining the variance of the voiced frames, determining the average of the current and past unvoiced frames, determining the variance of the current and past unvoiced frames, and determining the average of the determined voiced frames and 19. The method of claim 18, comprising determining the first merit value from a variance value and the average and variance values of the determined unvoiced frames. 21. The step of determining the first merit value, the step of obtaining the sum of the variance values, the step of calculating a weighted sum of the variance values, and the step of calculating the weighted sum of the variance values of the unvoiced frame from the average value of the voiced frames. A step of subtracting an average value, a step of squaring the subtracted value, a step of dividing the weighted sum of the variance values by the squared value, and a step of generating the first merit value from the division result. 21. The method of claim 20, comprising: 22. The step of calculating the weighted sum calculates the first probability that the first voiced sound detector indicates that the current frame is a voiced frame, and the first voiced sound detection. Calculating a second probability that the instrument indicates that the current frame is an unvoiced frame, the variance value of the current and past voiced frames being the first
22. The method of claim 21, comprising multiplying a probability and multiplying the variances of the current and past unvoiced frames by the second probability, and forming the weighted sum from the results of this multiplication. . 23. The method of claim 22, wherein the dividing step comprises multiplying the division result by the first and second probabilities to produce the first merit value.