JPH045393B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a word speech recognition method for recognizing words by comparing input speech with a word dictionary in which phonemes are expressed.

(Structure of Conventional Example and its Problems) FIG. 1 is a functional block diagram of an example of a conventional word speech recognition method and an apparatus for implementing an embodiment of the word speech recognition method of the present invention. Conventional example first
This will be explained with reference to FIGS. 2 and 3. In Figure 1, 1 is a parameter extraction unit that creates a time series of parameters from input speech, 2 is a probability density calculation unit that calculates the probability density of a phoneme by collating standard patterns of phonemes, and 3 is a segmentation unit for each phoneme. This is a word recognition unit that performs similarity calculations, likelihood calculations, and word similarity calculations. 4 is a phoneme standard pattern section that stores a phoneme standard pattern that represents the distribution of various parameters for each phoneme in the form of an average value (μ _i ) for each phoneme and a covariance matrix (Σ _i ) between various parameters. , 5 is a word dictionary section that stores a word dictionary in which all words to be recognized are expressed as symbol examples in phoneme units. In the word dictionary, for example, the words "Asahi" and "Syuhin" are written as "ASAHI" and "SJUHIN=".

Next, the operation of the above conventional example will be explained. The parameter extraction unit 1 analyzes input phonemes every 10 ms frame, extracts parameters, and creates a parameter time series. Next, probability density calculation section 2
In this step, the parameters obtained for each frame are compared with the phoneme standard pattern, and the probability density of the phoneme is calculated. Next, the word recognition unit 3 uses the above parameters and the obtained probability density value to perform phoneme segmentation for each phoneme according to the dictionary phoneme sequence that constitutes the dictionary item. Then, the type of phoneme and the likelihood l of the segmented interval corresponding to the phoneme are calculated according to the following formula, and the degree of similarity is determined as the average of the likelihoods of each phoneme in this dictionary entry. Here, let the phoneme be X, and the frame numbers at the start and end of the segmented section corresponding to X are N _s , N _e
If the value of each parameter in the n-th frame is 〓 _o , then the likelihood l _x of the phoneme X is defined by the following formula.

φ _i (〓 _o ) represents the probability density of a certain phoneme i,
It is defined as Eq.

φ _i (〓 _o )=1/(2π) ^N/2 ｜Σ _i ｜ ^1/2 exp
[-1/2 (〓 _o −〓 _i ) ^T-1 Σ _i (〓 _o −〓 _i )]... 〓 _o : N parameters (vector) in frame n 〓 _i : Parameter of a certain phoneme i Average value (vector) Σ _i : Covariance matrix In the equation, the range of the denominator summension i in probability density division differs depending on the phoneme X. For example, when X is the phoneme A, the range of i has five vowels: A, E, I, O, and U. The word similarity L _M obtained above was determined for each dictionary item according to the formula, and the dictionary item with the maximum L _M was selected as a recognized word.

L _M = _NP 〓 ^j=1 l _j /NP ... L _M : Similarity of the Mth word in the dictionary l _j : Likelihood of the jth phoneme in the dictionary phoneme sequence NP: Number of dictionary phonemes Figure 2 Each phoneme when saying /ASAHI/ (Asahi) /A/, /S/, /A/, /H/, /I/
It represents the time change of the probability densities φ _A , φ _S , φ _A , φ _H , and φ _I . In this case, the segmentation and likelihood calculation of each phoneme for the dictionary word /ASAHI/ are the dictionary phoneme series /A/, /S/, /A/, /
According to the order of H/, /I/, the first phoneme /
For A/, correspond the segmented interval (a-b) using φ _A , and according to the formula, φ _A
Calculate l _A using , and similarly find l _S , l _A , l _H , and l _I.

Figure 3 shows the phonemes /S/, /J/, /U/, /H/, / when uttering /SJUHIN=/ (guest of honor).
The probability density of I/, /N=/φ _S , φ _J , φ _U , φ _H , φ _I ,
It shows the time change of φ _N= . In Figure 3,
The segmentation of each phoneme for the dictionary word /SJUHIN=/ is the dictionary phoneme sequence /S/, /
This is done in the order of J/, /U/, /H/, /I/, /N=/, but the first phoneme /S/
When performing _{segmentation of} The probability densities φ _J and φ U of /J/ and /U/, which should originally have large values in the interval _/ and /U/, have smaller values than φ _S φ _H. Therefore, in the phoneme sequence, the first /S/
Because the original segmentation interval (c-d) is mistaken for (c-f), the segmentation of the second and subsequent phonemes is incorrect, and the likelihood becomes lower, resulting in Words that include voiceless consonants, persistent consonants, vowels that are likely to be voiced unvoiced or are pronounced slowly, and words that contain phoneme sequences with continuous consonants have the disadvantage that they are easily misrecognized.

(Objective of the Invention) The present invention is intended to eliminate the drawbacks of the conventional example described above, and aims to improve the accuracy of likelihood calculation, thereby improving the word recognition rate.

(Structure of the Invention) In order to achieve the above-mentioned object, the present invention, for a 4-consecutive phoneme sequence of a voiceless consonant, a persistent consonant, a devoiced vowel, or a vowel and a consonant whose occurrence is easily sluggish, the above-mentioned 4 continuous phonemes are grouped together. By performing segmentation and calculating the likelihood, it is possible to obtain the effect of improving the accuracy of the segmentation and likelihood calculation.

(Description of Embodiment) The following is an embodiment of the present invention shown in FIGS. 1 and 3.
This will be explained with figures. In FIG. 1, a parameter extraction section 1, a probability density calculation section 2, and a phoneme standard pattern section 4 have the same configuration and functions as those of the conventional example. The word dictionary stored in the word dictionary section 5 describes the words to be recognized as phoneme symbol strings, but the difference from the conventional example is that the words to be recognized are represented by voiceless consonants, persistent consonants, and vowels that are easy to be pronounced devoiced or half-voiced. A code is attached in advance to the four consecutive phoneme sequences of consonants and consonants.

The operation of this embodiment will be explained. First, a parameter extraction section 1 obtains parameters for each frame from the input speech, and a probability density calculation section 2 uses the values of the parameters to calculate probability densities obtained from each phoneme standard pattern. The process up to this point is the same as the conventional example. Next, the word recognition unit performs segmentation of the phoneme X for each dictionary entry according to the dictionary phoneme series that constitutes that dictionary entry, and then segments the phoneme X and the segmented interval corresponding to the phoneme X. To calculate _{the likelihood l}
For continuous phoneme sequences, since the probability density value of persistent consonants or vowels does not appear sufficiently, the above four continuous phonemes are segmented together, and the four continuous phonemes are segmented according to the formula.
Find the likelihood of continuous phonemes.

Here, in contrast to the formula, in the case of other ordinary phonemes, the likelihood is calculated using the formula as in the past.
In Figure 3, the probability density of each phoneme in the /SJUH/ portion changes over time until around the beginning of /U/.
The probability density φ _S of S/ is dominant, followed by the probability density φ _H of /H/ until near the beginning of /I/. Therefore from the movements of φ _S , φ _H and φ _I (c-g)
Let be the segmentation interval of /SJUH/,
The likelihood l _SJUH of /SJUH/ is calculated from the formula using φ _S and φ _H.

In this example, voiceless consonants, persistent consonants,
4. Vowels and consonants that are easily devoiced or unvoiced
Since segmentation and likelihood calculation are performed on continuous phoneme sequences, there is an advantage that the recognition rate of words containing two consecutive phonemes of vowels that are easily devoiced or unvoiced in dictionary words is improved.

(Effects of the Invention) The present invention has the above-described configuration, and provides the following effects.

When performing segmentation and likelihood calculation of phoneme sequences of voiceless consonants, persistent consonants, vowels and consonants that are easily devoiced or sluggish in pronunciation, four consecutive phonemes of voiceless consonants, obstruent consonants, vowels and consonants that are easy to be devoiced or sluggish in voicing are By performing segmentation and likelihood calculation all at once, there is an advantage that segmentation and likelihood calculation can be performed with higher accuracy than conventional methods.

[Brief explanation of drawings]

FIG. 1 is a diagram for explaining the word speech recognition method in the conventional method and an embodiment of the present invention, and FIG.
Each phoneme φ _A when uttering ASAHI/ (morning sun),
A diagram showing the time changes of φ _S , φ _A , φ _H , and φ _I. Figure 3 shows φ _S when uttering /SJuHIN=/(chief guest)
FIG. 3 is a diagram showing changes over time in φ _J , φ _U , φ _H , φ _I , and φ _N= . 1... Parameter extraction section, 2... Probability density calculation section, 3... Word recognition section, 4... Phoneme standard pattern section, 5... Word dictionary section.

Claims (1)

[Claims] 1. Word speech recognition that recognizes words in input speech using a word dictionary that describes the words to be recognized as symbol strings in phoneme units and standard patterns for each phoneme that are expressed as distributions of acoustic parameters for each phoneme. In the method, the input speech is matched with each dictionary entry of a word dictionary,
The input speech is segmented for each phoneme according to the dictionary phoneme sequence that constitutes each dictionary item, and the probability density that the segmented speech section is generated from that phoneme is determined by using the standard pattern of that phoneme. is calculated, and for the segmented speech interval, the similarity between each dictionary item and the input speech is determined using the above probability density value. A word speech recognition method characterized in that for a series of four consecutive phonemes of vowels or vowels and consonants that are easy to be uttered, the four consecutive phonemes are segmented together and likelihood calculation is performed.