JPS60202494A - Word voice recognition - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

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 its Problems) FIG. 1 is a functional block diagram of an example of a conventional word speech recognition method and an apparatus for executing an embodiment of the word speech recognition method of the present invention. Figure 1 shows a conventional example.

This will be explained with reference to FIGS. 2 and 3. In Figure 1,
1 is an i4 that creates a time series of motherboard meters from input audio.
2 is a probability density calculation unit that calculates the probability density of a phoneme by collating the phoneme standard number 4, and 3 is a word recognition unit that performs segmentation, likelihood calculation, and word similarity calculation for each phoneme. .

In addition, 4 is a phoneme standard that stores a phoneme standard pattern that represents the distribution of various parameters for each phoneme in the form of an average value (μi) for each phoneme and a covariance matrix (Σi) between various types of meters. The button part 5 is a word dictionary part 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, for example, the word "Sarapolo"
, "Kantei" is r 5AQPOROJ, r KA
It is written as N=NAI J, etc.

Next, the operation of the above conventional example will be explained. In the parameter extraction unit 1, the input phoneme is analyzed for each frame of 10m5, and the parameter is extracted to create a parameter time series. Next, the probability density calculation unit 2 compares the parameters obtained for each frame with the phoneme standard/yatan of the phoneme standard/motan unit 4, and calculates the probability density of the phoneme. next,
The word recognition unit 3 performs phoneme segmentation for each dictionary item according to the dictionary phoneme series that constitutes the dictionary item, and calculates the type of phoneme and the likelihood of the segmented interval corresponding to the phoneme according to the following formula 0. degree t
is calculated, and the similarity is calculated as the average of the likelihoods of each phoneme in that 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 /4' parameter in the nth frame is Cn, then The likelihood tX is defined by the following formula.

φ, (Cn) represents the probability density of a certain phoneme i, and is defined as in equation (2).

1) In Equation 0, the range of summation 1 in the denominator in the division calculation of the probability density differs depending on the phoneme X. For example, when X is the phoneme A (7), the range of i is 5 vowels,
A, E, 1.0. It is set as U.

The word similarity M obtained from the above was determined for each dictionary item according to formula (2), and the dictionary item with the maximum LM was selected as a recognized word.

LM=, 'kuretj/NP -■ J=1 Figure 2 shows the time change in the probability density of each phoneme in the /AN=NA/ part when uttering /KAN=NA I/ (cabinet). There is. In this case, the segmentation and likelihood calculation of the /AN=NA/ part are performed for each phoneme /A/,
/N=/, /N/, /A/ probability density value φ□
.

φ, -1φ8. Segmentation is performed according to the time change of φ. If /AN=NA/, the first /A
Correspond the segmented interval (a-h) to /, calculate LA using φ□ according to ■2, and /N
Similarly, tN= for =/, /N/, and /A/.

Calculate tN * tA.

FIG. 3 shows the temporal change in the probability density of each phoneme when the same unit yg /KAN = NA I / is uttered by different speakers. In Figure 3, the segmentation and likelihood calculation for /AN=NA/ are φ, φ, −1(5
) φ9. This is done based on the time change of φ, but when segmenting /N=/, the phoneme that comes after /N=//
The probability density φ, of N/ is not large enough in the /N/ interval, and φ,= has a large value in the /N/ interval, and the next phoneme /A
We have reached the beginning of the / section. Therefore, the segmentation interval of /N-/ becomes the interval (gh), which includes the interval of /N/, so the segmentation of the next phoneme /N/ after /N-/ is incorrect, and the likelihood tN is also low. Words containing two consecutive phonemes, such as a nasal or a nasal, had 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 sound repellent.

When segmenting and calculating the likelihood of a phoneme sequence with continuous nasal sounds, the accuracy of the segmentation and likelihood calculation can be improved by segmenting and calculating the likelihood of two consecutive phonemes of a nasal sound and a nasal sound at the same time. It is.

(6) (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 yatan is the same as the conventional example. The word dictionary represents the single @ to be recognized as a string of phoneme symbols. Moreover, the /ya parameter time series obtained by the /ya parameter extraction is the same as in the conventional example. The operation of this embodiment will be explained. First, the noise parameter extraction unit 1 obtains a parameter for each frame from the input audio,
Furthermore, the probability density calculation unit 2 calculates the value of the parameter and
Calculate the probability density obtained from each phoneme standard ij methane.

Next, in the word recognition unit 3, segmentation of the phoneme X is performed for each dictionary item in the word dictionary unit 5 according to the dictionary phoneme series that constitutes that dictionary item. The likelihood tx of the interval is calculated.
When there are two consecutive phoneme sequences of nasal sounds, the probability density value of the first phoneme, the phlegm, is dominant until the end of the next nasal sound. Therefore, two consecutive phonemes, a nasal and a nasal, are segmented together, and the likelihood is calculated for the segmented interval. Probability density φ6 of each phoneme /A/, /N-/, /N/, /ni/ in the /AN=NA/ part of Figure 3
．． φ, -1φ8. Looking at φ□, φ, - has a larger value than φ, at the /N/ portion, and continues until the beginning of /A/ (h). Therefore, using the value of φ, -, /N=N/
Segmentation is performed on the two consecutive phonemes from g to h, and the likelihood tN-N for the two phonemes is determined according to the formula 0 using the values of φ and - for the segmented interval (gh). Here, in contrast to the 0 formula, in the case of a normal phoneme, the 0 formula is used to calculate the likelihood as in the conventional case.

In this implementation, in order to perform segmentation and likelihood calculation by combining the phoneme sequences of phonics and nasals into one,
This method has the advantage of improving the recognition rate of words containing two consecutive phonemes, such as a nasal or a nasal.

However, the use of symbols follows the formulas ■ and ■.

(Effects of the Invention) As described above, the present invention can perform segmentation and likelihood calculation with higher precision than conventional methods by collectively segmenting two consecutive phonemes of a phonic and nasal sound and performing likelihood calculation.

[Brief explanation of the drawing]

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.
Probability density φ6 of each element /A/, /N=/, /N/, /A/ in the part /AN=NA/ when /(kannai) is uttered. φ, -9φ8. Figure 3 shows the change in φe over time.
If you say /, φeh, φ, -1φ,. It is a figure showing a time change of φ. 1... Parameter extraction unit 2... Probability density calculation unit, 3... Word recognition unit 4... Phoneme standard slam unit 5... Word dictionary unit (9) Diagram 1 Contains meeting voice complaint early wild boar

Claims (1)

[Claims] We perform word recognition of input speech using a word dictionary that describes the words to be recognized as symbol strings for each phoneme, and standard numbers for each phoneme that are represented by the distribution of the acoustic parameters of each phoneme. In the word speech recognition method, the input speech is checked against each dictionary entry in a word dictionary, the input speech is segmented for each phoneme according to the dictionary phoneme series that constitutes each dictionary entry, and the standard nomaton of that phoneme is used. The probability density that the segmented speech section is generated from the phoneme is calculated, and the similarity between each dictionary item and the input speech is calculated using the above probability density value for the segmented speech section. When recognizing words, the phonograph in the dictionary word,
A word speech recognition method characterized in that, for a phoneme sequence with continuous nasal sounds, two consecutive phonemes of a nasal sound and a nasal sound are segmented together and likelihood calculation is performed.