JPH045392B2 - - 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 apparatus for executing an example of a conventional word speech recognition method and an embodiment of the word speech recognition method of the present invention. Conventional example first
This will be explained with reference to FIG. 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 comparing standard patterns of phonemes, and 3 is a segmentation unit for each phoneme. This is a word recognition unit that performs calculations such as probability calculation, likelihood calculation, and word similarity calculation. In addition, 4 represents the distribution of various parameters for each phoneme, created in advance through preliminary experiments, etc., in the form of an average value for each phoneme (〓 _i ) and a covariance matrix (Σ _i ) between various parameters. A phoneme standard pattern section 5 stores phoneme standard patterns, and a word dictionary section 5 stores a word dictionary in which all words to be recognized are expressed in symbol strings in phoneme units. In the word dictionary, for example, the words "Satsuporo" and "Kurume" are written as "SAQPORO" and "KURUME", respectively.

Next, the operation of the above conventional example will be explained. The input phoneme is analyzed by the parameter extraction unit 1 every 10 ms frame, parameters are extracted, and a parameter time series is created. The probability density calculation unit 2 compares the parameters obtained for each frame with the phoneme standard pattern and calculates the probability density of the phoneme. Next, word recognition section 3
For each dictionary item, phoneme segmentation is performed for each phoneme according to the dictionary phoneme sequence that constitutes the dictionary item using the above parameters and the obtained probability density value, and the phoneme is segmented according to the following formula. , and the likelihood l of the segmented interval corresponding to that phoneme is calculated, and the 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 N _s and _Ne , and the value of each parameter in the nth frame is _Co , The likelihood l _x of X is defined by the following formula.

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

φ _i (C _o )=1/(2π) ^1/2 | Σ _i | ^1/2 exp [
−1/2 (C _o −〓 _i ) ^T ₋₁ 〓 _i (C _o −〓 _i )〕 ... _Co : N parameters (vector) in the n-th frame μ _i : Average of parameters of a certain phoneme i Value (vector) Σ _i : Covariance matrix In the formula, the range of summension i in the denominator in probability density division differs depending on the phoneme X. For example, when X is the phoneme A, The range is 5 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 _i : Likelihood of phoneme j in the dictionary phoneme sequence NP: Number of dictionary phonemes Figure 2 shows / Each phoneme when saying KURUME/ (Kurume) /K/, /U/, /R/, /U/, /
The probability densities of M/, /E/, φ _K , φ _U , φ _R , φ _U ,
It shows the changes in φ _M and φ _E over time. In this case, segmentation and likelihood calculation of each phoneme for the dictionary word /KURUME/ are performed using the dictionary phoneme sequence /
K/, /U/, /R/, /U/, /M/, /
E/, and the first phoneme /K/ is made to correspond to the interval (a-b) segmented using φ _K , and according to the formula, using φ _K , l _K
, and similarly obtain l _U , l _R , l _U , l _M , l _E .

Figure 3 shows the temporal change in the probability density of each phoneme when the same word /KURUME/ is uttered by different speakers. In Figure 3, the dictionary word /
The segmentation of each phoneme for KURUME/ is the phoneme series /K/, /U/, / in the dictionary.
Segmentation is performed according to the order of R/, /U/, /M/, /E/, but when segmenting the first phoneme /K/, the probability density φ _K of /K/
appears second in the phoneme sequence in the dictionary /U/
is dominant until near the beginning of , and in the section of /U/ that appears first in the phoneme sequence of the dictionary, φ _U is φ _K
It is a small value compared to . Further, in the section /R/, φ _R is also approximately the same value as φ _K.

Therefore, originally it should be (c-d) /K/
Since the interval is mistaken for interval (c-e) or interval (c-f), the likelihood of error in the segmentation of the second and subsequent phonemes is also lower, resulting in unvoiced consonants, unvoiced vowels, or unvoiced Words that contain phoneme sequences with continuous vowels or voiced consonants that are easy to be sloppy have the disadvantage of being 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 object, the present invention provides a method for segmentation and likelihood calculation of a devoiced vowel sandwiched between a voiceless consonant and a voiced consonant or a vowel whose pronunciation has been omitted. Alternatively, the accuracy of segmentation and likelihood calculation can be improved by segmenting and calculating the likelihood of consecutive three phonemes of voiceless consonants, vowels, and voiced consonants, including vowels that are half-voiced. It's something you get.

(Description of Embodiment) An embodiment of the present invention will be described below with reference to FIGS. 1 and 3. In FIG. 1, the phoneme standard pattern is the same as in the conventional example. In the word dictionary, the words to be recognized are represented by examples of phoneme symbols, but unlike the conventional example, codes are attached in advance to devoiced vowels or vowels that are likely to be pronounced half-heartedly. Further, the parameter time series obtained by parameter extraction is the same as in the conventional example.

The operation of this embodiment will be explained. First, parameters for each frame are obtained from the input speech, and then the probability density obtained from each phoneme standard pattern is calculated using the parameter values. The process up to this point is the same as the conventional example. Next, segmentation of the phoneme X is performed for each dictionary entry according to the dictionary phoneme series that constitutes that dictionary entry, and the _likelihood of the segmented interval corresponding to that phoneme X and that phoneme is calculated, but if there is a devoiced vowel sandwiched between a voiceless consonant C ₁ and a voiced consonant C ₂ or a vowel V that is easy to pronounce half-voiced in the dictionary phoneme sequence,
The value of the probability density of V does not indicate the nature of a vowel, but rather the nature of a voiceless or voiced consonant. Therefore, by using the type of each phoneme in the sequence of unvoiced consonants, unvoiced vowels, unvoiced vowels, and voiced consonants (C ₁ VC ₂ ), the phoneme, and the corresponding probability density value of each phoneme, Then, the three phonemes are segmented together, and the likelihood l _C1vc2 is calculated for the segmented interval.

In Figure 3, in the interval (d-e) of /U/ next to /K/, the probability degree φ _U of /U/ has almost no value, and instead the probability density φ _K of /K/ is the second
It is dominant up to the beginning of the /U/. Further, the probability density φ _R of /R/ is approximately the same value as φ _K in the interval of /R/. Therefore, the interval (c
−f) is the segmentation interval of the phoneme sequence /KUR/ in which /K/, /U/, and /R/ are combined into one, and the values of φ _K and φ _R within the segmentation interval of /KUR/ The likelihood l _KUR for the three phonemes is calculated using the formula.

In contrast to the formula, for other ordinary phonemes, likelihood calculations are performed using formulas as in the past.

In this example, the vowels that have been uttered are
Since segmentation and likelihood calculations are performed on unvoiced consonants, vowels with incomplete pronunciation, and phoneme sequences of unvoiced consonants, rather than treating them as single vowels, the recognition rate for words containing vowels with incorrect pronunciation is improved. be.

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

When segmenting and calculating the likelihood of a vowel with a half-voiced sound sandwiched between a voiceless consonant and a voiced consonant, the unvoiced consonant containing the half-voiced vowel, the half-voiced vowel, and the three consecutive phonemes of the unvoiced consonant are summarized. By performing segmentation and calculating the likelihood, this method has the advantage that the 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 when saying KURUME/ (Kurume) /K/, /U/, /R/, /U/, /M/, /
A diagram showing the temporal changes in the probability densities φ _K , φ _U , φ _R , φ _U , φ _M , φ _E of E/. Figure 3 shows the case in Figure 2 when a different speaker uttered /KURUME/ φ _K , φ _U , φ _R ,
FIG. 3 is a diagram showing changes over time in φ _U , φ _M , and φ _E ; 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 that performs single-error recognition of input speech using a word dictionary that describes the word to be recognized as a symbol string for each phoneme and a standard pattern for each phoneme that is expressed as a distribution of the acoustic parameters of each phoneme. In the recognition method, the input speech is checked against each dictionary item in the word dictionary,
Segment the input speech for each phoneme according to the dictionary phoneme sequence that constitutes each dictionary entry, and use the standard pattern of that phoneme to calculate the probability density that the segmented speech section is generated from that phoneme. When recognizing words by calculating the similarity between each dictionary entry and the input speech using the above probability density value for the segmented speech interval, For unvoiced vowels and vowels with half-voiced parts, group the continuous three phonemes of unvoiced vowels and vowels with half-voiced parts, unvoiced consonants including unvoiced vowels and half-voiced vowels, unvoiced vowels or vowels with half-voiced parts, and voiced consonants. A word speech recognition method characterized by performing segmentation and calculating likelihood.