JPH045396B2 - - 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.

(Constitution of Conventional Example and Problems thereof) FIG. 1 is a functional block diagram of an example of a conventional word speech recognition method and an apparatus for carrying out 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 compares phoneme standard patterns and calculates the probability density of phonemes, and 3 is a segmentation unit for each phoneme. , a word recognition unit that performs likelihood calculations and word similarity calculations. In addition, 4 is a phoneme standard pattern that stores a phoneme standard pattern that represents the distribution of various parameters for each phoneme in the form of an average value for each phoneme (〓 _i ) and a principal variance matrix (Σ _i ) between various parameters. Section 5 is a word dictionary section in which a word dictionary in which all words to be recognized are expressed in symbol strings in units of phonemes is stored. The word dictionary is "Asahi", "Suji" is "ASAHI", "SUZI"
etc. is written as.

Next, the operation of the above conventional example will be explained.

In parameter extraction section 1, the input phoneme is
Analyze every s frames, extract parameters, and create a parameter time series. Probability density calculation section 2
Then, the parameters obtained for each frame are compared with the phoneme standard pattern, and the probability density of the generated phoneme is calculated from the parameter values. Next, in the word recognition unit 3, using the above parameters and the obtained probability density, phoneme segmentation is performed for each phoneme according to the dictionary phoneme sequence that constitutes the dictionary item for each dictionary item, The likelihood l of the phoneme and the segmented interval corresponding to the phoneme is calculated according to the following formula, and the degree of similarity is determined as the average of the likelihoods of each phoneme in the dictionary entry. Here, if the phoneme is X, the frame numbers at the start and end of the segmented section corresponding to X are Ns, Ne, and the value of each parameter in the nth frame is Cn, then the likelihood of phoneme X is l _x is defined by the following formula.

φ _i represents the probability density of a certain phoneme i, and is defined as in the equation.

φ _i (〓n)=1/(2π) ^J/2 ｜Σ _i ｜ ^1/2 exp [−1/2(〓n−μ _i ) ^T Σ _i ^-1 (〓n−μ _i )]...
〓n: J parameters in the n-th frame (vector) μ _i : Average value of the parameters of a certain phoneme i (vector) Σ _i : Covariance matrix In the formula, the range of the submension i of the denominator in probability density division is: It depends on what the phoneme X is. For example, when X is the phoneme A (a), i
The range is 5 vowels A, E, I, O, and U.
From the above, the obtained word similarity L _M is calculated for each dictionary item according to the formula, and the dictionary entry with the maximum L _M is recognized as a word. L _M = _NP 〓 ^k=1 l _k /NP... L _M : Similarity of the M-th word in the dictionary l _k : Likelihood of the k-th phoneme in the dictionary phoneme series NP: Number of dictionary phonemes Interposed between consonants C ₁ and C ₂ in the dictionary phoneme series , when there is a devoiced vowel or a vowel V that is easy to pronounce in a half-voiced manner, the type of each phoneme in the consonant C ₁ , the devoiced vowel or a vowel V that is _half- voiced, the /C ₁ VC ₂ / arrangement of the consonant C 2 and Using the probability density value of each phoneme corresponding to the phoneme arrangement, segmentation is performed for the three phonemes at once, and the likelihood lc ₁ vc ₂ is calculated for the segmented interval. FIG. 2 shows the temporal change in the probability density of each phoneme /S/, /U/, /Z/, and /I/ when /SUZI/ is uttered. In Figure 2, /
In the interval (b-c) of U/, there is almost no probability density φ _U of /U/, but instead, the probability density φ _S of /S/ in the first half of /U/, and the probability density φ _Z of /Z/ in the second half. It is dominant.

Also, φ _Z is dominant until the beginning of /I/ (d), and from (d) on, φ _I becomes dominant. Therefore, the interval (a-d) is /
Let the segmentation interval of the phoneme sequence /SUZ/ which combines S/, /U/, /Z/ into one, /
In the segmentation interval (a-d) of SUZ/, using the values of φ _S and φ _Z , the likelihood l _SUZ
Calculate.

FIG. 3 shows the temporal changes in the phonemes /Z/, /U/, /S/, and /I/ when uttering /ZUSI/. In this case, the /U/ interval (f
-g), there is almost no probability density φ _U of /U/, and instead, the probability density φ _Z of /Z/ is dominant in the first half of /U/, and the probability density φ _S of /S/ is dominant in the second half. Also, φ _S is dominant until the beginning of /I/ (h), and from (h) on, φ _I becomes dominant. Therefore, the interval (eh)
, /Z/, /U/, /S/ are combined into one segmentation interval of the phoneme sequence /ZUS/, and in the segmentation interval (e-h) of /ZUS/, φ _Z , The likelihood l _ZUS is calculated according to the formula using the value of φ _S. In the formula, the probability density of the first phoneme C ₁ and the third phoneme C ₂ of three consecutive phonemes are calculated. The order in which the probability densities appear is not considered. Therefore, the second
In Fig. 3, the intervals (a-d) and (e-h) for three phonemes satisfy the condition of Eq. In other words, in the example of /SUZ/ and /ZUS/, if the frame sums of φ _Z and φ _S in an interval of three phonemes are almost the same, then the likelihood of /SUZ/ l _SUZ and /
Likelihood of ZUS/l There is no difference in scores for _ZUS . Therefore, as a result, two words containing a phoneme sequence in which two consonants are replaced in a three-consecutive phoneme sequence of a consonant, a devoiced vowel, or a half-voiced vowel, and a consonant, such as /SUZ/ and /ZUS/. It had a defect that made it difficult to identify.

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

(Structure of the Invention) In order to achieve the above object, the present invention provides consonants,
When segmenting three consecutive phonemes of devoiced vowels, half-voiced vowels, and consonants together and performing likelihood calculations, within the segmented interval, the part that shows the characteristics of the first consonant and ,
Detect one or more boundaries of the part showing the characteristics of the second consonant, and within the segmented interval,
For each boundary, based on the likelihood obtained from the part showing the characteristics of the first consonant and the part showing the characteristics of the second consonant, we calculate the likelihood of summarizing the three consecutive phonemes. The accuracy of the likelihood calculation is improved by calculating the boundary and using the maximum likelihood therein as the likelihood of that section.

(Description of Embodiment) An embodiment of the present invention will be described below with reference to the drawings. In FIG. 1, the phoneme standard pattern is the same as in the conventional example. In a word dictionary, words to be recognized are expressed as phoneme symbol strings, and vowels that are likely to be devoiced or uttered slowly are given codes in advance. Further, the parameter time series obtained by parameter extraction is the same as in the conventional example.

Next, the operation of the above embodiment will be explained. First, a parameter extractor 1 obtains parameters for each frame from the input speech, and a probability density calculator 2 uses the values of the parameters to calculate the probability density obtained from each phoneme standard pattern. In the word recognition unit 13, segmentation of the phoneme X is performed for each dictionary item according to the dictionary phoneme series that constitutes the dictionary item, and the phoneme X and the likelihood of the segmented interval corresponding to the phoneme X are segmented. When calculating _the degree _l
and _{the type of each phoneme in the sequence C 1 VC 2} / of the consonant C 2, the phoneme, and the corresponding probability density value of each phoneme. Calculate the likelihood l _C1VC2 for the segmented interval. The process up to this point is the same as the conventional example. Here, the likelihood
When calculating l _C1VC2 , the likelihood l _C1 (α) of the part where the probability density of the first phoneme C ₁ is dominant, and the likelihood of the part where the probability density of _{the first phoneme C 1 is} dominant, Calculate the likelihood l _C2 (α) of the part where the probability density of consonant C ₂ is dominant,
Find l _C1VC2 according to the formula. In Figure 2, /
In the segmentation interval (a-d) of SUZ/,
When calculating the likelihood l _SUZ , the probability density φ _S of /S/ and /
Using the probability density φ _Z of Z/, the first consonant /
Find one or more boundaries between the S/ part and the second consonant /Z/ (α=i, b, ha).

Find the likelihood l _S (α) of the first consonant part obtained using φ _S and the likelihood l _Z (α) of the second consonant part obtained using φ _Z , and _s (α) and l _z (α)
Based on , calculate the likelihood l _SUZ (α) using the formula.

However, Na: frame number of boundary α Ns, Ne: start and end frame numbers in the segmentation section of 3 consecutive phoneme series /C ₁ VC ₂ / φ _C1 , φ _C2 : 3 consecutive phoneme series /C ₁ VC ₂ / Probability density of phoneme C ₁ and phoneme C ₂ in , l _C1VC2 = Max (l _C1VC2 (α), α = 1 to k) (k is the number of boundaries) ... sum of l _C1 (α) and l _C2 (α) If the likelihood is l _C1VC2 (α), then α is (a), (b), and (c) in Figure 2, so
It becomes like the formula.

l _SUZ (A) | _s (A) + l _Z (A) l _SUZ (B) | _s (B) + l _Z (B) l _SUZ (C) | _s (C) + l _Z (C) ...... As a result, l _SUZ (a), l _SUZ (b), l _SUZ (c) are calculated respectively, and according to the formula α=a, b, c, l _SUZ (a), l _SUZ (b), l Let the one with the highest numerical value among _SUZ (c) be the likelihood l _SUZ of the three consecutive phonemes /SUZ/. By using the above-described likelihood calculation method, the two words /SUZI/ and /ZUSI/ can be easily identified.

In the above embodiment, the likelihood of a 3 consecutive phoneme sequence of a consonant, a devoiced vowel, or an unvoiced vowel, and a consonant can be determined with high accuracy.
This method has the advantage that two words including a phoneme sequence in which two consonants in a continuous phoneme sequence are exchanged can be easily recognized.

(Effects of the Invention) The present invention segments a consonant, a silent vowel or a vowel with a half-voiced sound, and three continuous phonemes of a consonant together, and when performing likelihood calculation, the part that shows the characteristics of the first consonant. By finding one or more boundaries of the part that shows the characteristics of the second consonant, calculating the likelihood for each, and setting the maximum likelihood as the likelihood of that interval, this method is more accurate than conventional methods. It has the advantage of being able to perform likelihood calculations.

[Brief explanation of drawings]

FIG. 1 is a diagram for explaining a word speech recognition method according to the conventional method and an embodiment of the present invention, and FIG.
Each phoneme /S/, / when saying SUZI/
Probability densities of U/, /Z/, /I/ φ _S , φ _U , φ _Z ,
Figure 3 shows the time change of φ _I , and each phoneme /Z/, /U/, / when uttering /ZUSI/ is shown in Figure 3.
FIG. 3 is a diagram showing temporal changes in the probability densities φ _Z , φ _U , φ _S , and φ _I of S/ and /I/. 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. Compare input speech with each dictionary item of a word dictionary,
The input speech is segmented for each phoneme according to the dictionary phoneme series that constitutes each dictionary item, and the distance between the standard pattern of the phoneme and the phoneme is used for the segmented phoneme interval to create each dictionary entry. In a word speech recognition method that recognizes words by determining the similarity between items and input speech, for three consecutive phonemes of consonants, devoiced vowels, or half-voiced vowels, and consonants, the three consecutive phonemes are segmented together. Then, within the segmented interval, one or more boundaries between the part showing the characteristics of the first consonant and the part showing the characteristics of the second consonant of the three phonemes are detected, and the segmentation is performed. Within the interval, for each boundary, the likelihood obtained from the part showing the characteristics of the first consonant and the part showing the characteristics of the second consonant, the likelihood that summarizes the three consecutive phonemes. is calculated for each boundary, and the maximum likelihood among them is taken as the likelihood of that section. 2. As a standard pattern for each phoneme, use a standard pattern expressed by the distribution shape of the acoustic parameters of each phoneme,
Word speech recognition according to claim 1, characterized in that the probability density that a segmented speech section is generated from the phoneme is used as a distance measure between the standard pattern of phonemes and the phoneme. Method.