JPH05281994A - Similarity arithmetic unit - Google Patents

Abstract

PURPOSE:To obtain the similarity arithmetic unit which is small in arithmetic quantity without deteriorating the category decision making performance for a speech signal. CONSTITUTION:A probability density distribution storage means 2 is stored with plural probability density distributions showing acoustic features of the same category and a 2nd category similarity arithmetic means 3 finds the 2nd category similarity 9 of an input signal by using the specific probability density distribution 8a; and a probability density distribution selecting means 4 selects the probability density distributions, category by category, according to the 2nd category similarity 9 and a 1st category similarity arithmetic means 5 performs similarity arithmetic for the specified probability density distribution 8b and outputs category similarity 11.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a voice recognition device for recognizing a voice by using a category similarity for each voice signal partial section.

[0002]

2. Description of the Related Art FIG. 8 is a block diagram showing the configuration of a conventional similarity calculating device as shown in "Voice Recognition by Stochastic Model" by Seiichi Nakagawa (published by The Institute of Electronics, Information and Communication Engineers, P72).

In the figure, reference numeral 41 is a sound analysis means for acoustically analyzing a voice signal for each constant section and converting it into an acoustic parameter vector, and 42 is a probability density distribution storage means for storing a probability density distribution representing suitable acoustic characteristics of a category. , 43 is a category similarity calculation means for calculating the similarity of the acoustic parameter vector with respect to the stored probability density distribution and outputting the category similarity, 44 is a voice signal, 45 is an acoustic parameter vector, 46 is a probability density distribution, 47 is a category similarity.

The operation of the conventional similarity calculating device will be described below by taking as an example the case of using a probability density distribution consisting of a mixed continuous distribution having a probability density distribution of 8. FIG. 9 is a flowchart showing the operation of the conventional similarity calculation device.

In FIG. 9, x is an acoustic parameter vector, p is a category number, θ8 (p) is a probability density distribution representing a category p consisting of a mixed continuous distribution having a distribution number of 8, b
(Θ8 (p), x) is the similarity of the acoustic parameter vector x to the probability density distribution θ8 (p), B (p) is the category similarity of the category p, and N is the total number of categories.

In FIG. 9, the process starts from start 301 and ends at end 307. As shown in FIG. 10, it is assumed that a probability density distribution θ8 (p) representing acoustic characteristics of each category is stored in advance in the probability density distribution storage means 42 prior to the similarity calculation.

First, at 302, the acoustic analysis means 41 performs acoustic analysis on the input voice signal 44 to convert it into an acoustic parameter vector x. Next, steps 303 to 3
In 06, the category similarity calculation means 43 repeats the similarity calculation for the category p of the acoustic parameter vector 45 output from the acoustic analysis means N times, which is the total number of categories, for the probability density distribution θ8 (p) of the category p. The similarity B (p) is output. This category similarity B
By (p), it is possible to determine which category the input audio signal was, and it is possible to perform the Viterbi operation in the HMM using the category similarity as the output probability.

[0008]

The conventional similarity calculating device is configured as described above, and since the similarity calculation for a specific probability density distribution is always repeated regardless of the input signal, the continuous mixing with a large number of distributions is performed. There is a problem that the amount of calculation becomes large when using a probability density distribution consisting of distributions.

The present invention has been made to solve the above problems, and an object of the present invention is to obtain a similarity calculation device having a small amount of calculation without degrading the category discrimination performance of a voice signal.

[0010]

In addition to the configuration of the conventional similarity calculating device, the similarity calculating device according to the present invention has a probability density storing a plurality of kinds of probability density distributions representing acoustic characteristics of voices belonging to the same category. A second category that calculates the category similarity of the acoustic parameter vector using the distribution storage means and the specific probability density distribution stored in the probability density distribution storage means for each category, and outputs it as the second category similarity. Similarity calculation means, probability density distribution selection means for selecting the probability density distribution of the probability density distribution storage means for each category according to the second category similarity, and outputting a probability density distribution designation signal, and the acoustic analysis means. Category probability calculation is performed using the probability density distribution specified by the probability density distribution specifying signal of the acoustic parameter vector output from The first is obtained by a structure comprising a category similarity calculating means for.

[0011]

The similarity calculating device according to the present invention determines the neighborhood category of the input signal based on the second category similarity, and uses the probability density distribution having a higher category decision ability for the neighborhood category which is important in category decision. It is possible to reduce the amount of computation without degrading the category determination ability by performing the similarity computation by using the probability density distribution with a smaller computation amount for the non-neighborhood category. Become.

[0012]

【Example】

Example 1. An embodiment of the present invention will be described below with reference to FIG. In FIG. 1, reference numeral 1 is an acoustic analysis means for acoustically analyzing a voice signal at regular intervals and converting it into an acoustic parameter vector. The storage means 3 calculates the category similarity of the acoustic parameter vector using the specific probability density distribution stored in the probability density distribution storage means 2 for each category, and outputs it as the second category similarity. It is a category similarity calculation means.

Numeral 4 is a probability density distribution selecting means for selecting a probability density distribution for each category according to the second category similarity and outputting a probability density distribution designating signal. Reference numeral 5 is a probability density designated by the probability density distribution designating signal. First category similarity calculation means for calculating and outputting the category similarity of the acoustic parameter vector output from the acoustic analysis means using distribution, 6 is a voice signal, 7a and 7b are acoustic parameter vectors, and 8a and 8b are probabilities. The density distribution, 9 is the second category similarity, 10 is the probability density distribution designation signal, and 11 is the category similarity.

Hereinafter, in the first category similarity calculating means, two kinds of probability density distributions, which are mixed continuous distributions of distribution number 8 and distribution number 2, are selected and used as probability density distributions.
The operation of the similarity calculation device according to the present invention will be described by taking as an example the case where a probability density distribution formed of a mixed continuous distribution with a distribution number of 1 is used in the category similarity calculation means.

Originally, a mixed continuous distribution with a distribution number of 1 is a single continuous distribution, but here it is called a mixed continuous distribution with a distribution number of 1 for the sake of explanation. The probability density distribution consisting of the mixture distribution with the number of distributions 2 has a low category judgment ability, but the amount of calculation is small.
The probability density distribution, which is a mixture distribution of, has a high category judgment ability but a large amount of calculation.

FIG. 2 is a flow chart showing the operation of the similarity calculating device according to the present invention. In FIG. 2, x is an acoustic parameter vector, p is a category number, r (p, x) is a second category similarity with respect to the category p of the acoustic parameter vector x, T is a threshold value of similarity used for selecting the number of distributions, and K is Constant used to select the number of distributions,

Θ8 (p) is a probability density distribution representing a category p having a continuous distribution with a distribution number of 8, and θ2 (p) is a probability density distribution representing a category p having a continuous continuous distribution having a distribution number of 2. , Θ1 (p) is a probability density distribution that represents a category p consisting of a mixed continuous distribution with the number of distributions being 1, and b (θ8
(p), x) is the similarity of the acoustic parameter vector x to the probability density distribution θ8 (p), b (θ2 (p), x) is the similarity to the probability density distribution θ2 (p) of the acoustic parameter vector x, b (Θ1 (p), x) is the similarity of the acoustic parameter vector x to the probability density distribution θ1 (p), s (p) is the probability density distribution designation signal of category p, and B (p) is the category similarity of category p. , N is the total number of categories. Also,

[0018]

[Equation 1]

Indicates that a value obtained by subtracting the constant K from the maximum similarity R (p) is used as the threshold T.

In FIG. 2, the process starts at start 101 and ends at end 120. Prior to the similarity calculation, as shown in FIG. 3, the probability density distribution storage means 2 previously stores the probability density distributions θ1 (p) and θ2 (p) representing the acoustic features of each category.
And θ8 (p) are stored. First, at 102, the acoustic analysis means 1 performs acoustic analysis on the input voice signal 6 to convert it into an acoustic parameter vector x.

Next, in steps 103 to 106, the second
The category similarity calculation means 3 calculates and outputs the category similarity R (p) using the probability density distribution θ1 (p). In step 107, the probability density distribution selecting means 4 determines the threshold value T of the category similarity, and in steps 108 to 113, the neighborhood category whose category similarity is larger than the threshold T is set as the probability density distribution designation signal s (p). 8 is given, and the probability density distribution designation signal s (p) is set to 2 for a non-neighboring category whose category similarity is equal to or lower than the threshold T.

Next, in steps 114 to 119, the first
The category similarity B in the category similarity calculation means 5
(p) is calculated. When the probability density distribution designation signal s (p) is 8, the probability density distribution θ8 (p) is calculated in step 116.
If the probability density distribution designating signal s (p) is 2, the category similarity B (p) is calculated using the probability density distribution θ2 (p) in step 117. This similarity calculation is repeated N times, which is the total number of categories, and the category similarity B (p) is output. It is possible to determine which category the voice signal input is based on the category similarity B (p). Further, the Viterbi operation in the HMM can be performed by using the category similarity as the output probability.

Example 2. Another embodiment based on the present invention will be described below with reference to FIG. In FIG. 4, reference numeral 21 is an acoustic analysis means for acoustically analyzing a voice signal at regular intervals and converting it into an acoustic parameter vector, and 22 is a probability density distribution storing a plurality of kinds of probability density distributions representing acoustic characteristics of voices belonging to the same category. The storage means 23 calculates the category similarity of the acoustic parameter vector using the specific probability density distribution stored in the probability density distribution storage means for each category, and outputs it as the second category similarity.
It is a category similarity calculation means.

Reference numeral 24 is a probability density distribution selecting means for selecting a probability density distribution for each category according to the second category similarity and outputting a probability density distribution designating signal, and 25 is a probability density designating by the probability density distribution designating signal. The category similarity of the acoustic parameter vector output from the acoustic analysis unit is calculated and output using the distribution, or the second category similarity output from the second similarity calculation unit is input and output as the category similarity. The first category similarity calculating means for selecting whether or not by the probability density distribution designating signal, 26 is a voice signal, 27a and 27b are acoustic parameter vectors, 28
a and 28b are probability density distributions, 29a and 29b are second category similarities, 30 is a probability density distribution designation signal, and 31 is a category similarity.

Hereinafter, in the second category similarity calculating means, two kinds of continuous continuous distributions having a distribution number of 1 and a distribution number of 8 are used as the probability density distribution, and in the first category similarity calculating means, the second category similarity calculating means. The operation of the similarity calculation device according to the present invention will be described by taking as an example the case of using a mixed distribution having a distribution number of 1, which is common to. The probability density distribution composed of the mixture distribution with the number of distribution 1 has a low category determination ability but a small amount of calculation, and the probability density distribution composed of the mixture distribution with a distribution number of 8 has a high category determination capability but a large amount of computation.

FIG. 5 is a flow chart showing the operation of the similarity calculation device in the second embodiment. In FIG. 5, x is an acoustic parameter vector, p is a category number, r (p, x) is the similarity of the acoustic parameter vector x to the category p,
T is a threshold value of similarity used for selecting the number of distributions, K is a constant used for selecting the number of distributions,

Θ8 (p) is a probability density distribution representing a category p having a continuous distribution with a distribution number of 8, and θ1 (p) is a probability density distribution representing a category p having a continuous continuous distribution having a distribution number of 1. , B (θ8 (p), x) is the similarity of the acoustic parameter vector x to the probability density distribution θ8 (p), b (θ1
(p), x) is the similarity of the acoustic parameter vector x to the probability density distribution θ1 (p), s (p) is the probability density distribution designating signal of the category p, B (p) is the category similarity of the category p, N Is the total number of categories. Also,

[0028]

[Equation 2]

Indicates that a value obtained by subtracting the constant K from the maximum similarity R (p) is used as the threshold T.

In FIG. 5, the process starts at start 201 and ends at end 220. Prior to the similarity calculation, as shown in FIG. 6, the probability density distribution storage means 22 stores the probability density distributions θ1 (p) and θ8 in advance representing the acoustic features of each category.
It is assumed that (p) is stored. First, step 202
At the sound analysis means 21, the input voice signal 26
Then, acoustic analysis is performed on the above and converted into an acoustic parameter vector x.

Next, in steps 203 to 206, the second
In the category similarity calculation means 23, the probability density distribution θ1 (p)
Is calculated, and the similarity is output as the category similarity R (p). In step 207, the probability density distribution selecting means 24 determines the threshold value T of the category similarity, and in steps 208 to 213, the neighborhood category whose category similarity is larger than the threshold T is set as the probability density distribution designation signal s (p). 8 is given, and the probability density distribution designation signal s (p) is set to 1 for the non-neighboring category whose category similarity is less than or equal to the threshold T.

Next, in steps 214 to 219, the first
In the category probability calculation means 25, the category similarity B
(p) is calculated. When the probability density distribution designating signal s (p) is 8, the probability density distribution θ8 (p) is used to calculate the category similarity B (p) at 216, and the probability density distribution designating signal s (p) is calculated.
When (p) is 1, the similarity calculation using the probability density distribution θ1 (p) is not performed, and the category similarity R (p) output from the second category similarity calculation means in step 217.
Is the category similarity B (p).

As a result, the similarity calculation using the probability density distribution θ1 (p) can be reduced and the amount of calculation can be reduced. This similarity calculation is repeated N times, which is the total number of categories, and the category similarity B (p) is output. It is possible to determine which category the voice signal input is based on the category similarity B (p). Also, this category similarity is used as the output probability H
Viterbi operation in MM can be performed.

Example 3. Another embodiment based on the present invention will be described below with reference to FIG. In FIG. 7, reference numeral 61 is an acoustic analysis means for acoustically analyzing a voice signal at regular intervals and converting it into an acoustic parameter vector, and 62 is a probability density for storing a plurality of types of probability density distributions representing acoustic characteristics of voices belonging to the same category. The distribution storage means 63 calculates the category similarity of the acoustic parameter vector using the specific probability density distribution stored in the probability density distribution storage means for each category, and outputs it as the second category similarity. It is a category similarity calculation means.

64 is a probability density distribution selecting means for selecting a probability density distribution for each category according to the second category similarity by defining a selection criterion by a selection control signal input from the outside, and outputting a probability density distribution designating signal. Reference numeral 65 denotes a category similarity of the acoustic parameter vector output from the acoustic analysis unit, which is calculated by using the probability density distribution designated by the probability density distribution designation signal, and is output, or is output from the second similarity calculation unit. The first category similarity calculation means selects whether to input the two-category similarity and output it as the category similarity according to the probability density distribution designation signal.

66 is a voice signal, 67a and 67b are acoustic parameter vectors, 68a and 68b are probability density distributions, 6
Reference numerals 9a and 69b are second category similarities, 70 is a probability density distribution designation signal, 71 is a category similarity, and 72 is a selection reference control signal.

Similarly to the second embodiment, the second category similarity calculating means uses two kinds of continuous continuous distributions having a distribution number of 1 and a distribution number of 8 as the probability density distribution, and the first category similarity calculating means As an example, the case of using a mixture distribution having a distribution number of 1, which is common to the second category similarity calculation means,
The operation of the similarity calculation device according to the present invention will be described.

The acoustic analysis circuit 61, the probability density distribution storage means 62, the second category similarity calculation means 63 in FIG.
The operation of the first category similarity calculation means 65 is the same as the corresponding means of the second embodiment.

Probability density distribution selecting means 6 in this embodiment
4 also uses the probability density distribution selection signal 6 as an evaluation criterion whether or not the second category similarity is larger than the threshold T, as in the second embodiment.
4, the value of the constant K used to obtain the threshold value T can be changed by the selection reference control signal 72 externally input.

If the value of K is made smaller, the threshold value T becomes larger and the number of operations for the probability density distribution θ8 (p), which has a large amount of calculation, becomes smaller, so that the amount of calculation of the similarity calculating device becomes smaller, and conversely K
The larger the value, the larger the amount of calculation. In this way, it becomes possible to control the calculation amount of the similarity calculation device by the selection reference control signal 72 input from the outside.

In the above embodiment, the probability density distribution having a high category judgment ability is a probability density distribution consisting of a mixed continuous distribution having a distribution number of 8, and the probability density distribution having a small calculation amount has a distribution number of 1 or 2. Although the case where a probability density distribution consisting of mixed distributions of is used has been described as an example, the number of distributions may be another number, or a multidimensional Gaussian distribution with a high category determination ability, a probability with a small amount of calculation As the density distribution, different types of probability density distributions such as a diagonal Gaussian distribution may be combined and used, and the same effect is obtained.

In the above embodiment, the probability density distribution expressing the same category in the first category similarity calculating means has been described as two kinds, but it may be three kinds, four kinds, or more kinds. The same effect is achieved.

In the above description, all categories have been described as having their own probability density distributions, but a plurality of categories are represented by a common probability density distribution, and this is calculated as the second category similarity calculation. When it is used in the means, an even larger amount of calculation reduction effect is produced.

[0044]

As described above, according to the present invention, a probability density distribution storing means for storing a plurality of kinds of probability density distributions representing acoustic characteristics of voices belonging to the same category, and the probability density distributions for each category. Second similarity calculation means for calculating the category similarity of the acoustic parameter vector using a specific probability density distribution stored in the storage means and outputting it as the second category similarity, and according to the second category similarity The probability density distribution selecting means for selecting the probability density distribution of the probability density distribution storage means for each category and outputting the probability density distribution designating signal, and the sound using the probability density distribution designated by the probability density distribution designating signal Since the first similarity calculation unit that calculates and outputs the category similarity of the acoustic parameter vector output from the analysis unit is provided, the input signal The neighborhood category is judged by the category similarity, and the similarity calculation is performed on the probability density distribution with a higher category judgment ability for the neighbor categories that are important in category judgment. By performing the similarity calculation on the density distribution, there is an effect that a similarity calculation device having a small calculation amount of the similarity calculation can be realized without deteriorating the category determination capability.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing a similarity calculation device according to a first embodiment of the present invention.

FIG. 2 is a flowchart showing the operation of the similarity calculation device according to the first embodiment of the present invention.

FIG. 3 is an explanatory diagram showing a probability density distribution stored in a probability density distribution storage unit according to the first embodiment of the present invention.

FIG. 4 is a configuration diagram showing a similarity calculation device according to a second embodiment of the present invention.

FIG. 5 is a flowchart showing the operation of the similarity calculation device according to the second embodiment of the present invention.

FIG. 6 is an explanatory diagram showing a probability density distribution stored in a probability density distribution storage means according to the second embodiment of the present invention.

FIG. 7 is a configuration diagram showing a similarity calculation device according to a third embodiment of the present invention.

FIG. 8 is a block diagram showing a conventional similarity calculation device.

FIG. 9 is a flowchart showing the operation of a conventional similarity calculation device.

FIG. 10 is an explanatory diagram showing a probability density distribution stored in a conventional probability density distribution storage means.

[Explanation of symbols]

 1 Acoustic analysis means 2 Probability density distribution storage means 3 Second category similarity calculation means 4 Probability density distribution selection means 5 First category similarity calculation means 6 Speech signals 7a, 7b Acoustic parameter vectors 8a, 8b Probability density distribution 9 Second Category similarity 10 Probability density distribution designation signal 11 Category similarity 21 Acoustic analysis means 22 Probability density distribution storage means 23 Second category similarity calculation means 24 Probability density distribution selection means 25 First category similarity calculation means 26 Speech signal 27a, 27b Acoustic parameter vector 28a, 28b Probability density distribution 29a, 29b Second category similarity degree 30 Probability density distribution designation signal 31 Category similarity degree 61 Acoustic analysis means 62 Probability density distribution storage means 63 Second category similarity degree calculation means 64 Probability density distribution Selection means 65 1st category similarity performance Means 66 the speech signal 67a, 67b acoustic parameter vector 68a, 68b probability density distribution 69a, 69b second category similarity 70 probability density distribution designation signal 71 category similarity 72 selection criteria control signal

Claims (3)

[Claims] 1. A similarity calculation device for calculating a category similarity of a voice signal by using a probability density distribution that represents the acoustic characteristics of the voice for each recognition category, and acoustically analyzes the voice signal at regular intervals to obtain an acoustic parameter vector. The acoustic analysis means for converting into a probability density distribution storage means for storing a plurality of kinds of probability density distributions representing acoustic characteristics of voices belonging to the same category, and the probability density distribution storage means for each category. A second category similarity calculating means for calculating the category similarity of the acoustic parameter vector using a specific probability density distribution and outputting it as the second category similarity;
The probability density distribution selecting means for selecting the probability density distribution of the probability density distribution storing means for each category according to the second category similarity and outputting the probability density distribution designating signal, and the probability density distribution designating signal are designated. The first category similarity calculation means for calculating and outputting the category similarity of the acoustic parameter vector output from the acoustic analysis means using the probability density distribution is provided, and the probability density is calculated for each category according to the input voice signal. A similarity calculation device characterized by selecting a distribution type and calculating a category similarity. 2. The first category similarity calculation means calculates the category similarity for a designated probability density distribution and outputs the category similarity, or the second similarity calculation means outputs the second category similarity. The similarity degree computing apparatus according to claim 1, wherein the category similarity degree is inputted and is output as a category similarity degree, which is selected by the probability density distribution designation signal. 3. The configuration according to claim 1, wherein the probability density distribution selecting means is configured to change a criterion for selecting a probability density distribution according to a selection criterion control signal input from the outside. Similarity calculator.