JPS5876894A - Monosyllable recognition equipment - 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 The present invention relates to an improvement in a speech recognition device that recognizes monosyllables corresponding to Japanese character abbreviations that are uttered in segments.

Conventionally, there is a method that uses bang matching to identify single syllables that are uttered in sections.

Typical methods include the following.

Normally, in Japanese monosyllables, tsukitondo is in the form of a consonant and a vowel, so a method is used to recognize them separately. First, the input voice is analyzed and the consonant and vowel parts of the voice bang are extracted.

The vowel category is determined by matching the vowel sound part, that is, the vowel sound, with a pre-registered vowel standard sound.

Next, a consonant category is determined by matching the consonant button with a pre-registered consonant standard button, and a monosyllable recognition result is determined.

Generally, stable bangs can be obtained for vowel parts, so the cutting method and batan matching method are different from the methods of 1. There are various transformations. For example, one method is to
One possible method is to cut out a number of p frames of voice bangs and use them as consonant bangs. Since this method compares patterns with a fixed number of frames, there is no need to expand or contract the time axis. Therefore, recognition can be performed with a small amount of calculation.

However, the duration of consonants may vary depending on the pronunciation or category.

If the number of frames at this time is adjusted to the length of the time minus the length of the consonant, a large amount of memory will be required to store the standard bang. Furthermore, in this method, when comparing consonants with short durations, the rounding feature of comparing using a long number of frames may be weakened. On the other hand, if the number of frames is adjusted to the length of a consonant with a short duration, it becomes difficult to recognize a consonant with a long duration.

A second method is to cut out the consonant part according to the length of the consonant, and then use a dynamic programming method to match the consonant part and the standard consonant part by expanding and contracting the time axis non-linearly. Using this method,
Buttons of different lengths can be mapped to each other with a fine grain of mi <0. However, this dynamic programming method requires a large amount of calculation. Furthermore, unlike the case of word recognition, in the case of single syllable recognition, the effect of nonlinear expansion and contraction of the time axis is considered to be small, so it cannot necessarily be said to be the optimal method.

An object of the present invention is to recognize monosyllables with a small amount of standard memory and a small amount of calculation, and to obtain high recognition performance. In order to achieve this object, the monosyllable recognition device according to the present invention includes an analysis unit that converts nine monosyllables that are uttered into segments into a sound slam, and extracts a consonant part and a vowel part from the sound sound, and converts the consonant sound into a sound sound. and a voice extraction unit that generates a vowel bang; a standard bang memory unit that stores monosyllable consonant bangs and vowel bangs that have been uttered in advance as consonant standard bangs and vowel standard bangs, respectively; In the vowel and pa sound matching section, which matches vowel standard drums and determines the vowel category, when matching the input consonant drum and the above-mentioned consonant standard drum, if the time lengths of the two are different, a vowel is added after the shorter consonant drum. The consonant matching section includes a consonant matching section that performs matching by adding a bang.

An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing the overall structure of an embodiment according to the present invention. ! The voice input from Iku Sho 7 on 1 is analyzed by the analysis section 2 and output as a voice bang P. Consonant bang C and vowel bang V are cut out in the voice cutting section 3 from the voice beat P. When registering standard patterns, these are saved in the consonant standard button memory 4 and the vowel standard button memory 5, respectively. held within. At the time of recognition, first the vowel slam V is ψ in the vowel matching section 6, and the vowel standard bang ■R and ma are matched.

For this matching method, find the distance between the vowel bangs≠. With this k) /a/, /i, /: /lshi/
It recognizes vowel categories such as e/; 10/, and outputs the result as a vowel category VC. Consonant matching 117 performs matching of consonant parts and converts the recognition result into /kl/v'
etc. are output as consonant category CC. The operating principle of the consonant matching section 7 will be explained. FIG. 2 is a conceptual diagram for explaining an example of the operation of the consonant matching unit 7. Input consonant bang 11 (the figure is an example of a 5 frame bang)
Suppose that we want to find the distance between the consonant standard I tongue 12 (the figure shows an example of eight frames of tongue). Each audio pattern is a time series of vectors with the time direction pointing to the right. From the beginning of the voice to the fifth frame, as shown by arrow 13, the frames of the input consonant bang and the consonant standard slam at the same position are compared to find the distance. From the 5th frame onwards, the consonant standard button 12 is changed to 1 of the human-generated voice.
The vowel button 15 for one frame is made to correspond as shown by the arrow 14, and the distance is determined. The consonant bang is usually followed by the vowel bang, and the vowel bang is almost constant over time. Therefore, by continuing a short crawling consonant bang with a vowel bang, the original bang can be reproduced.

With this method, it is possible to easily match battens of different lengths. Furthermore, since it is sufficient to separately store the consonant part and vowel part drums as needed as the standard drums, the amount of memory required for the standard drums can be reduced. The example shown here shows a case where the input consonant bang is shorter than the standard consonant bang, but in the opposite case, just replace both and follow the same procedure as above. Figure 3 2 is a block diagram showing a specific example of the configuration of a consonant matching section 70 circuit. FIG. Consonant bang C of the input voice
is a vowel slam in the consonant buffer 21 V is a vowel buffer 22
is maintained. First, the consonant standard slam CR is held in the consonant standard tap buffer 23, and the vowel standard tap VB is held in the vowel standard tap buffer 24. The input bang frame counter 25 outputs a consonant buffer 721 K frame address lAt-. A consonant pattern CF for each frame is output from the consonant buffer 721 in accordance with this address. Similarly, the standard slam 7 frame counter 26 is a child as
The frame is addressed to the s pattern buffer 23. This input slam 7 frame counter 25 and l1ls slam 7
The frame counter 26 synchronously counts up from 1, and when the value reaches a value equal to each frame number, the counter that has reached the equal value stops counting. Data selectors 27 and 28 select consonant bang CF and consonant standard bat'/ while the 7-frame counter continues counting.
cRF is selected and output to the distance calculation section 29. The distance calculation unit 29 calculates the distance between two bangs for each frame. The distance for each frame is 30
The distance for one bang is accumulated.

A case where the number of frames of the consonant bang is smaller than the standard consonant bang will be explained. Input button 7 frame counter 25
When the value reaches the number of consonant bang frames and counting stops, the data selector 27 selects the vowel frame data VF from the inverted vowel bang buffer 22 and outputs it to the distance calculation section 29. Standard slam frame counter 2
When the value of 6 reaches the number of 7 rays of the consonant standard bang, it means that the distance between both methane has been found, and the distance cumulative value d becomes 7.
The above is a case where the number of frames of the input consonant bang is smaller, but it operates in the same way when the number of frames of the counter-pressure consonant standard bang is small. A vowel standard bang VRF is added after the consonant standard slam CHP. This distance integrated value d is input to the minimum value calculation unit 31 for each standard slam category. Here we calculate the minimum value in all consonant categories,
The consonant category CC that takes the minimum value is output as a recognition result.

The embodiments described above are mere examples selected for convenience of explanation, and the present invention is not limited to these embodiments.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing one embodiment of the present invention, FIG. 2 is a conceptual diagram for explaining an example of the operation of the consonant matching section, and FIG. 3 is an example of a specific circuit of the consonant matching section. FIG.

Claims (1)

[Claims] A segmentation section that converts the single syllables that have been uttered into sections into bangs and makes them sound bangs, a voice cutting section that cuts out the consonant parts and vowel parts from the sound padding, and makes consonant bangs and vowel bangs; A standard bang memory section stores consonant bangs and vowel bangs of monosyllables that have just been uttered as consonant standard bangs and vowel standard bangs, respectively, the input vowel bangs, and the vowel standard bangs! , a vowel matching unit that determines the vowel category, and when matching the input consonant stamp and the standard consonant stamp, if the time lengths of the two are different, the vowel matching section adds the vowel stamp after the shorter consonant stamp. A monosyllable recognition device comprising a consonant matching unit for matching.