JPH045399B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to a 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 block diagram showing an outline of the functions of a device that implements one conventional word recognition method (first conventional example), and Fig. FIG. 6 is a diagram showing a change in the ratio of the middle band power to the high band and low band power in the J interval.

In FIG. 1, 1 is a parameter extraction section, 2 is a phoneme segmentation section, 3 is a phoneme recognition section, and 4 is a phoneme recognition section.
5 is a word dictionary section, 5 is a confusion matrix section, and 6 is a word recognition section. The word dictionary section 4 stores a word dictionary in which all words to be recognized are expressed in phonemes, and the word dictionary includes, for example, the words "Satsuporo", "Asahikawa", "Gyoda", "Zoda", etc. , "SAQPORO", "ASAHIKAWA",
It is written as "GJOODA", "ZOODA", etc.
The confusion matrix unit 5 shows the probability that various phonemes used in dictionary notation will be recognized in actual phoneme recognition, for example, the probability that A will be recognized as A is 85%, and the probability that A will be recognized as 0 is 85%. The probability that A will be dropped from the recognized phoneme sequence due to a segmentation error is 5%, and so on.

Next, the operation of the above conventional example will be explained. The parameter extraction unit 1 analyzes the input audio every 10ms frame and extracts the parameters.The phoneme segmentation unit 2 performs phoneme segmentation (input audio The phoneme recognition unit 3 performs phoneme recognition for each segmented interval. Word recognition unit 6
Now, the degree of similarity between the recognized phoneme sequence obtained by the phoneme recognition unit 3 and each dictionary item stored in the word dictionary unit 6 is calculated using the confusion matrix in the confusion matrix unit 5, The word (dictionary entry) with the highest degree of similarity was selected as the recognized word. The segmentation method for the Y vowel (hereinafter referred to as J) in this conventional example is based on the band power P M in the middle range (600 to 1500 Hz) of the input voice, the band power P _M in the high range (1500 Hz)
~4000Hz) band power P _H and low frequency (250~600Hz)
The ratio R _MM R _MM △=P _M ² /P _H ×P _L of the band power P _L (Hz) shown in the formula below is divided into segments with the valley-shaped section as the section J as shown in Figure 2. It was something to look forward to.
This utilizes the property that the _RMM changes in a valley shape in response to changes in the first and second formant frequencies in the J interval, but it has the following drawbacks. In other words, since R _MM is a quantity that fluctuates depending on the change in formant frequency even outside the J interval, the valley of R _MM occurs even in intervals other than the J interval, and in order not to mistake that interval as the J interval, the depth of the valley of R _MM is A certain threshold was set, and when the depth of a valley exceeded that threshold, that valley was segmented as section J, and other shallow valleys were ignored. As a result, only about 70% of the J sections were correctly segmented, and there were many erroneous word recognitions. For example, in Figure 1, the input word is GJOODA.
, this J is often not segmented for the reasons mentioned above, and in the example shown in Figure 1, the phoneme recognition result for G is incorrect as B, and the resulting recognized phoneme sequence becomes BOODA. As a result of calculating the similarity with each dictionary item in the word dictionary, the word recognition result was incorrectly ZOODA. This conventional example has a drawback in that there are many erroneous word recognitions as shown above.

Next, a second conventional example will be described with reference to FIG. In FIG. 3, the parameter extraction section 1 and word dictionary section 4 are the same as those in the first conventional example shown in FIG. In the second conventional example, the parameter extraction unit 1
The input section is analyzed and parameters extracted every 10 ms frame, and the word dictionary section 7 directly compares the parameter time series obtained in the parameter extraction section 1 with the word dictionary, and extracts the input voice for each dictionary item. Segment one phoneme at a time according to the dictionary phoneme sequence of the dictionary item, assuming that the person who uttered the dictionary item is the one who uttered the dictionary item, and the segmented interval is generated from that phoneme in the dictionary phoneme sequence. Words are recognized by calculating the likelihood, which is a measure of the certainty of a word, and determining the degree of similarity between the dictionary entry and the input speech as the average value of the likelihood. Here, in the second conventional example, when segmenting the J section corresponding to J on the dictionary phoneme sequence, the vowel whose distance is the closest to each vowel standard pattern obtained for each frame is segmented. The J section was segmented by utilizing the property that the type changes from I to A for "ya", from I to U for "yu", and from I to O for "yo".
However, in the second conventional example, unless the utterance is very clear, the characteristics of I are often not obvious in the J interval, and in that case, segmentation of the J interval is not possible. This tends to result in a judgment of "does not include", leading to word misrecognition.

(Object of the Invention) The present invention is intended to eliminate the drawbacks of the conventional example, and aims to improve the word recognition rate by making it possible to correctly segment the J section.

(Structure of the Invention) The speech recognition method according to the present invention compares the input speech with each dictionary item of a word dictionary in which the word to be recognized is expressed in phonemes, and inputs each phoneme according to the dictionary phoneme series that constitutes each dictionary item. In addition to segmenting the voice, when segmenting the Y-line semi-vowel, the ratio of the middle band power P _M of the input voice to the band powers P _H and P _L of the high and low frequencies R _ML △ = P _M ² / The temporal change of P _H ×P _L is also compared, and when the ratio R _ML is the minimum, segmentation is performed as a y-line semi-vowel. After that, each segmented speech segment corresponds to that phoneme. It calculates the likelihood, which is a measure of the probability that something has occurred, and uses this likelihood value to determine the degree of similarity between each dictionary entry and the input speech, and recognizes the words in the input speech. do. Due to this feature, the present invention can perform segmentation reliably and obtain a high word recognition rate.

(Description of Embodiment) An embodiment of the present invention will be described below with reference to the drawings. The basic configuration of the device used to carry out the method of this example is the same as that of the second conventional example.
This is illustrated by the block diagram in FIG. In FIG. 3, the parameter extraction section 1 and the word dictionary section 4 are the same as those in the first conventional example and the second conventional example.

The operation of this embodiment will be explained. First, the parameter extraction unit 1 analyzes the input voice every 10 ms frame and extracts parameters, and the word recognition unit 7 directly compares the parameter time series obtained by the parameter extraction unit 1 with the word dictionary of the word dictionary unit 4. For each dictionary item, assuming that the input speech is the one that uttered that dictionary item, segmentation is performed one phoneme at a time according to the dictionary phoneme sequence of that dictionary item, and the segmented interval is The word is calculated by calculating the likelihood, which is a measure of the probability that it was generated from that phoneme in the dictionary phoneme series, and calculating the degree of similarity between the dictionary entry and the input speech as the average value of the likelihood. Recognize. At this time, in this embodiment, when performing segmentation of the J section corresponding to J on the dictionary phoneme sequence, the mid-range band power of the input voice and the high-range and low-range bands shown in the above equation are used. Power ratio R _MM
However, as shown in FIG. 2, segmentation is performed using the valley-shaped section as the J section.

In this embodiment, even if a valley in R _MM occurs in a place other than the J section, there is no problem at all as long as J segmentation of dictionary entries different from the input word is not performed in that part. Even if such an incorrect segmentation is performed, it can be distinguished from the correct segmentation of the J interval by performing a likelihood calculation using other parameters so that the likelihood is low when calculating the likelihood. . Therefore, even if the valley of R _MM is used for segmentation of section J,
Unlike the first conventional example, there is no need to set a threshold for the depth of the valley, and even if the _RMM valley is shallow, the J section can be correctly segmented, and the word recognition rate is improved. Also, since R _MM rarely does not form a valley in section J,
Compared to the second conventional example, the rate at which the J section can be correctly segmented is much higher, and the word recognition rate is also improved. For example, as shown in Figure 3, when the input word is GJOODA, the segmentation of J is ensured,
The word recognition result was also correct: GJOODA.

(Effects of the Invention) The present invention has the above-described configuration, and provides the following effects. When the input speech is segmented according to the dictionary phoneme sequence of each dictionary item, the segmentation of J is expressed as the mid-range band power P _M of the input speech, and the high-range and low-range band powers P _H , P By taking advantage of the fact that the ratio of _L R _MM △=P _M ² /P _H ×P _L forms a valley in the J section, the J section can be reliably segmented, improving the word recognition rate. It has the advantage of

[Brief explanation of drawings]

Fig. 1 is a block diagram showing the outline of the functions of the device used to implement the first conventional speech recognition method, and Fig. 2 shows the mid-range band power and the high- and low-range bands in the J section. FIG. 3, which is a diagram showing changes in the power ratio in section J, is a block diagram showing an outline of the functions of the device used to implement the speech recognition method in the second conventional example and the embodiment of the present invention. be. 1...Parameter extraction section, 4...Word dictionary section,
7...Word recognition section.

Claims (1)

[Scope of Claims] 1. The input speech is compared with each dictionary entry of a word dictionary in which the word to be recognized is expressed in phonemes, and the input speech is segmented for each phoneme according to the dictionary phoneme series that constitutes each dictionary entry. When segmenting the Y-line semi-vowel, the ratio of the mid-range band power P _M of the input voice to the high- and low-range band powers P _H , P _L (R _MM △ = P _M ² /P _H ×P _L ), and when the ratio R _MM is the minimum, it is segmented as a ya semivowel, and then each segmented speech segment is the one in which that phoneme was uttered. A speech recognition method characterized by calculating a likelihood, which is a measure of certain certainty, and using this likelihood value to determine the degree of similarity between each dictionary item and input speech, and recognizing words in the input speech. .