JPS617894A - 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 speech recognition method for recognizing words by comparing input speech with a dictionary of words expressed in phoneme form.

Conventional configuration and its problems A conventional speech recognition method will be explained with reference to the drawings.

In the figure, the word dictionary represents all the words to be recognized using phonemes, such as Asahi and Yuki.

Yuki is /ASAH,I/, /JUUKI/, /JUKI
It is written as / etc. The phoneme standard, C-tan, is created in advance for each phoneme through preliminary experiments.

Next, the operation of the above-mentioned conventional example will be explained. The input voice is analyzed for each IQIIIsO frame, parameters are extracted and a parameter time series is created, and parameters are calculated in advance. Next, the degree of similarity is calculated for each dictionary item. When calculating this degree of similarity, the phoneme is segmented according to the dictionary phoneme sequence that makes up the dictionary item, and the segmented speech interval is used to utter the phoneme. The similarity is calculated as the average value of the likelihood of each phoneme in the dictionary entry, and the dictionary entry with the maximum similarity is selected as the recognized word.

Here, the long vowel, for example /UU/ in /JUUI<I/, is
Since it is usually not possible to find the boundary between /U/ and /[工/, segmentation is performed by grouping /UU/ together, and the likelihood is also calculated by grouping them together. In addition, the above /UU/
It is possible to think of a long vowel as one long vowel, but in this conventional example, it is two phonemes /U.
It is treated as if it is followed by /. Therefore, the likelihood value for two phonemes is also calculated.

In this conventional example, the likelihood of the second phoneme in the dictionary phoneme sequence! , is expressed by the following formula.・, e otsu = stone,
. -Ai, ・・・・・・■Here -e
□0 is calculated as a measure representing how well the parameters in each frame in the segmented interval match the standard pattern of that phoneme. Also,
□2 represents a reduction in likelihood when the segmented section is too long or too short. For long vowels, as mentioned above, the likelihood is calculated by grouping multiple phonemes together, so it is slightly different from formula 0, but it is basically the same, and the likelihood is determined by the degree of match with the standard vowel and the deduction of points depending on the length. do. Here, the words /JUUKI/ and /JUKI/ differ only in whether /U/ is a long vowel or a short vowel. In order to identify such words, in this conventional example, a long vowel, in this example /UU/
When calculating the likelihood of /UU/, if the length of the interval is shorter than a predetermined threshold, the likelihood is deducted, whereas for ordinary short vowels.

In this example, when calculating the likelihood of /U/, if the length of the interval of /'U/ is longer than the threshold, points are deducted from the likelihood.

However, the above conventional example had the following drawbacks. When the input word is /JUUKI/ and the dictionary phoneme sequence is also /JUUKI/, in the segmentation of /J/, the part with large parameter variation is divided into /J/
and /U/, but it is often the case that the /J/ section becomes extremely long and the /UU/ section becomes correspondingly short. This is because when a semi-vowel is followed by a long vowel, the parameter changes peculiar to semi-vowels last longer than when a short vowel follows. In other words, audibly, /JUU
/UU/ of / is clearly longer than /U/ of / J U /, but /UU/ of / J' U U / is
It is not simply an extension of /U/, but the properties of /J/ continue for a long time. Therefore, if we perform the segmentation of /J/ as described above, /JUU
The length of /J/ compared to /JU/ is taken up by the length of /J/, and /UU/ is obtained by removing /, T/.
This happens to be short for a long vowel. In such a case, for the input /J UUK T /, the likelihood of /U U / in dictionary entry /J
Since the / U / of UKI/ is not deducted, the likelihood is high and the degree of @ similarity is also higher for /JUKI/, which has the drawback that the word recognition result is incorrectly written as /JUKI/.

OBJECTS OF THE INVENTION The present invention eliminates the drawbacks of the above-mentioned conventional examples, and aims to improve the accuracy of similarity calculation, thereby improving the word recognition rate.

Structure of the Invention In order to achieve the above-mentioned object, the present invention provides the following steps when determining the appropriateness of the length of a long vowel or a short vowel following a semi-vowel.
By using the sum of the interval length of the long vowel or its short vowel and the interval length of the preceding semi-vowel, even if the interval length of the long vowel is short, if the sum of the interval length of the preceding semi-vowel is long, then the long vowel is short. On the other hand, even if the interval length of a short vowel is short, if the sum with the interval of the preceding semi-vowel is long, the likelihood is deducted. It has the effect of improving

゛Description of an Embodiment An embodiment of the present invention will be described below with reference to the drawings. The basic configuration of this embodiment is the same as that of the conventional example,
Further, the word dictionary and the standard pattern of phonemes are the same as in the conventional example.

The operation of this embodiment will be explained below. The difference between this embodiment and the conventional example is the method of calculating the likelihood of a vowel following a semi-vowel, and since the other parts are the same, the method of calculating the likelihood will be described below. Note that when we say semi-vowels here, we include not only semi-vowels at the beginning of words or between vowels, but also the semi-vowel part of the ink sound, ie /J/ such as riya (/RJA/). In the likelihood calculation, as in the conventional example, the input speech is segmented according to the dictionary phoneme sequence, and the likelihood is calculated using the 0 formula or a formula similar to the 0 formula in the case of a long vowel. However, in this embodiment, the deduction of the likelihood, that is, the method of determining pIp in the 0 equation is different from the conventional example. In this example, in calculating the likelihood of a long vowel following a semi-vowel, the deduction for being too short for a long vowel is not based on the length of the section containing only the long vowel, but on the difference in length between the long vowel section and the preceding semi-vowel. A threshold is set for the sum of , and this is done when the sum is σ shorter than the threshold. Furthermore, when calculating the likelihood of a short vowel following a semi-vowel, the point deduction for a short vowel that is too long is not based on the length of the segment containing only that vowel, but rather on the sum of the lengths of the short vowel segment and the preceding semi-vowel segment. A threshold is set for , and this is done when the sum is longer than the threshold.

An example of the effect of this embodiment will be described. The case of the input word /JUUKI/ shown in the figure, which is the same as the conventional example, will be described. In this input example, when the dictionary entry is also /JUUKI/, the segmentation result is the same as in the previous example, where the section for /J/ is long, and the section for /UU/ is short for each long ' sound. . However, in this example, since we look at the sum of the lengths of the /UU/ section plus the /J/ section, here the /J/ section is long, so the likelihood of being too short for /UU/ is deducted. There was no. Also, for the same human power /JUUKI/, if the dictionary entry /JUKI10, the segmentation result is /J/ is the same interval as /J/ in /J UUK I /, and /U/ is /UU/ in /J UUK I /. It was the same area. The length of kotte /U/ is the short vowel /U/
It was a standard length. However, in this example, the sum of the /U/ interval plus the /J/ interval is looked at, so the /J/ interval is very long, so the sum exceeds the threshold, and the /U/ interval is looked at. Likelihood was too long and points were deducted. As a result, for this /J UUK I / input,
The degree of similarity in the dictionary entry /JUUKI/ is greater than that in the conventional example because /UU/ is not deducted for being too short, while the degree of similarity in the dictionary item 7'JUI<I/ is greater than that in the conventional example. Compared to the case, /U/ is reduced by the point deducted for being too long, and the word recognition result is correct due to /U/.
JUUKI/Natsuta.

In this example, the likelihood deduction points for a long vowel that follows a semi-vowel that is too short, and the likelihood deduction points for a short vowel that follows a semi-vowel that is too long are calculated based on the length of the interval of those vowels and the interval of the preceding semi-vowel. By setting a threshold for the sum of s, it is possible to accurately deduct points for long vowels and short vowels that follow a semi-vowel, thereby improving the accuracy of likelihood calculation.

Effects of the Invention The present invention has the above-mentioned configuration, and limits the length of a long vowel that follows a semi-vowel, and limits the length of a short vowel that follows a semi-vowel. This is done by setting a threshold for the sum of the lengths plus the length of the interval and subtracting the likelihood, improving the accuracy of similarity calculation in the direction of increasing the degree of discrimination between long and short vowels.
A high word recognition rate can be obtained.

[Brief explanation of drawings]

The figure is a diagram showing a speech recognition method according to a conventional method and an embodiment of the present invention.

Claims (2)

[Claims] (1) Have a word dictionary in which the words to be recognized are expressed in phonemes, and a standard pattern for each phoneme, and match the input speech with each dictionary entry in the word dictionary, and then Segment the input speech for each phoneme, calculate the likelihood, which is a measure of the probability that the segmented speech interval is the one that uttered that phoneme, and use this likelihood value to identify each dictionary item. When recognizing an input word based on the similarity of the input speech, a restriction on whether a long vowel following a semi-vowel is too short is applied to the sum of the lengths of the long vowel and the preceding semi-vowel. voice recognition method. (2) The speech recognition method according to claim 1, wherein the short vowel following a semi-vowel is restricted from being too long with respect to the sum of the lengths of the short vowel and the preceding semi-vowel.