JPH0431117B2 - - 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.

(Constitution of Conventional Example and Problems thereof) FIG. 1 is a functional block diagram of an apparatus for executing an example of a conventional speech recognition method and an embodiment of the speech recognition method of the present invention. A conventional example will be explained with reference to FIG. In Fig. 1, 1 is a parameter extraction unit that creates a time series of parameters from input speech, 2 is a word recognition unit that performs segmentation for each phoneme, likelihood calculation, similarity calculation, etc., and 3 is a total number of words to be recognized. This is a word dictionary section that stores a word dictionary in which words are expressed as symbol strings in units of phonemes. In the word dictionary, for example, the words "Satsuporo", "Tokyo", "Toyonaka", and "Yamagata" are respectively "SAQPORO" and "TOOKYOO".
It is written as "TOYONAKA", "JAMAGTA", etc.

Next, the operation of the above conventional example will be explained. First, the input audio is analyzed in every 10 ms frame by the parameter extractor 1, parameters are extracted, and a parameter time series is created. The parameter time series is one in which parameters commonly used in subsequent processing are calculated in advance. Next, the word recognition unit 2 compares the word dictionary unit 3 to determine the degree of similarity for each dictionary item. When calculating this degree of similarity, the phoneme segment is determined according to the dictionary phoneme sequence that constitutes the dictionary item. The likelihood is calculated as a measure of the probability that the segmented speech interval is the one that uttered the phoneme, and the similarity is calculated as the average of the likelihoods of each phoneme in the dictionary entry. The dictionary entry with the highest degree of similarity is selected as the recognized word. Specifically, performing segmentation for a certain phoneme means that [(frame number of the rear end of the phoneme before the phoneme) + 1] is set as the start frame of the phoneme, and from there, the rear end frame of the phoneme is segmented. It is about searching and finding. The time length of the segmented speech segment for a certain phoneme is naturally within a certain range as long as the speech is natural. Therefore, when searching for the last frame of the phoneme, only a limited range is required. In this conventional example, this range is from 1 to
It was set at 30 frames (10 to 300 ms), but this value was appropriate for actual speech recognition.

However, the above conventional example had the following drawbacks. An example of this will be explained in conjunction with FIG. FIG. 2 is a diagram showing the correspondence of segmentation results between dictionary items TOJONAKA and JAMAGATA when the input voice is TOJONAKA (Toyonaka), with time oriented toward the right. In this example, the segmentation for the dictionary entry TOJONAKA was correct. on the other hand
Segmentation in the case of JAMAGATA is
There were two incorrect correspondences, TOJ-J and A-AGA, but in the likelihood calculation, even if the TOJ part of the input was considered as J, there was no problem on the parameters, and it was segmented as G. The section is A
Since this is an unstable part of the vocalization that transitions from
The degree of similarity also increased. For this reason, the input voice shown in this example was erroneously recognized as JAMAGATA. In the segmentation for the dictionary item JAMAGATA shown in this example, the section segmented with G was two frames, and the section segmented with the next A was only one frame. When segmenting a certain phoneme, if the time length of the segment is as short as 1 or 2 frames, the characteristics of that phoneme are weak in the utterance, and the transition between that phoneme and the next phoneme is similar to that of the next phoneme. It is often segmented into phoneme intervals, and therefore,
It is actually impossible for segmentations of short time duration to be continuous. Therefore, in this conventional method, as in the JAMAGATA example shown in Figure 2,
Although the method performed segmentation that would not be possible in reality, the degree of similarity increased, resulting in words being misrecognized.

(Objective of the Invention) The present invention eliminates the drawbacks of the conventional example described above, and aims to eliminate the clearly impossible segmentation as described above, thereby improving the word recognition rate.

(Structure of the Invention) The present invention collates input speech with each dictionary entry in a word dictionary, and segments the input speech for each phoneme according to the dictionary phoneme series that constitutes each dictionary entry.
The likelihood, which is a measure of the certainty that the segmented speech interval is the one that uttered the phoneme, is calculated, and this likelihood value is used to calculate the degree of similarity between each dictionary item and the input speech. In recognizing an input word, in order to achieve the above purpose, in phoneme segmentation, the time length of the interval of the phoneme is obtained by adding the time length of one or more previous phonemes. This method limits the time length of two or more phonemes to be too long or too short, eliminates clearly incorrect segmentation, and obtains the effect of obtaining a high word recognition rate.

(Description of Embodiment) An embodiment of the invention will be described below with reference to the drawings. The basic configuration of an apparatus for carrying out the method of this embodiment is shown in FIG. 1, as in the conventional example. In FIG. 1, the word dictionary is the same as in the conventional example.

The operation of this embodiment will be explained. First, the parameter extraction unit 1 analyzes input audio every 10 ms frame, extracts parameters, and creates a parameter time series. The process up to this point is the same as the conventional example. Next, this is compared with the word dictionary in the word dictionary section 2, and phoneme segmentation is performed for each dictionary item according to the dictionary phoneme series that constitutes that dictionary item. Here, in this embodiment, the range for searching for the end of a certain phoneme is limited to 1 to 30 frames as in the conventional case, and at the same time, in conjunction with the time length of the segmented section for the previous phoneme, The time length of two phonemes is limited to a certain range. For example, in the case of GA, the range is 5 to 44 frames. After segmentation, likelihood calculation is performed to obtain similarity as in the conventional method.

The effects of this embodiment will be described with examples. In the same input as the conventional example shown in FIG.
When the dictionary entry was TOJONAKA, segmentation was performed normally as in the conventional example. When the dictionary entry was JAMAGATA, the segmentation from the beginning of the word to G was the same as before, but G
Since the length of A is 2 frames, the rear end of the next A is searched within the range where the length of A is 3 to 30 frames, and segmentation as in the conventional method is not performed. In this example, the range to search for the trailing edge of A is
It was found that the segmentation of A was impossible, and it was determined that JAMAGATA could not be an input word. As a result, the input was correctly recognized as TOJONAKA. As described above, this embodiment has the advantage of being able to reduce misrecognition of words by eliminating clearly incorrect segmentations.

Although this embodiment shows an example of an input word in which only one word is uttered, the same effect can be obtained for continuous words or words in a sentence.

The present invention has the above-described configuration, and provides the following effects.

During phoneme segmentation, for the time length of two or more phonemes obtained by adding the time length of the phoneme interval to the time length of one or more previous phonemes. , too long or too short, and by limiting the rear end position of the phoneme, we can eliminate incorrect segmentation that is actually impossible, reduce misrecognition of words, and The recognition rate can be improved.

[Brief explanation of drawings]

FIG. 1 is a diagram showing the basic configuration of an apparatus for implementing a speech recognition method in a conventional example and an embodiment of the present invention. FIG. 2 is an explanatory diagram of segmentation in a conventional example. 1...Parameter extraction unit, 2...Word recognition unit,
3...Word dictionary section.

Claims (1)

[Claims] 1. The input speech is compared with each dictionary entry in 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. The likelihood, which is a measure of the probability that the segmented speech segment is the one that uttered the phoneme, is calculated, and this likelihood value is used to determine the degree of similarity between each dictionary entry and the input speech. When recognizing an input word, during phoneme segmentation, two or more phonemes are obtained by adding the time length of the interval of that phoneme to the time length of one or more previous phonemes. A speech recognition method characterized by limiting the time length of a phoneme to be too long or too short, and limiting the rear end position of the phoneme.