JPS62217297A - Word voice 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] [Summary] In a word speech recognition device, a pseudo word speech pattern created by concatenating monosyllable patterns is registered, and
For a specific word, a specific word audio pattern created based on the actual spoken word audio is registered, and the input word audio pattern created from the input word audio is used as at least one of the registered word audio patterns. Perform word recognition by comparing with As a result, better recognition performance can be obtained with less registration work.

[Industrial application field]

The present invention provides a word speech recognition device that recognizes input word speech by comparing word speech patterns registered in advance with input word speech patterns created from input word speech.
This invention relates to a word speech recognition device that has been improved so that a better recognition rate can be obtained with less registration work.

[Conventional technology] When recognizing unknown human-generated word speech, a pre-registered word pattern is compared with an input word speech pattern created from the input word speech, and the word with the smallest distance is selected as the recognition result. Word speech recognition devices that output words are often used.

FIG. 3 is a block diagram showing the basic configuration of a conventional word speech recognition device in which the word group to be recognized is not limited.

When unknown input ijl edict speech is input to the speech analysis section 210, the speech analysis section 210 analyzes the input word speech, extracts its characteristic parameters, detects sections, etc., and creates an input word speech pattern.

On the other hand, in the word pattern dictionary section 220, each word sound pattern of a word group to be recognized is registered in advance.

The word recognition section 230 uses the input word speech pattern inputted from the speech analysis section 210 and the word pattern dictionary section 220. The words with the shortest distance are compared with the speech patterns of each word read out, and the word with the smallest distance is output as the recognition result.

In such a conventional word speech recognition device in which the double type of the recognition target is not limited, the word pattern dictionary section 220
There are, for example, the following two methods for creating word patterns to be registered.

■ Speak each word in advance to create a word sound pattern to be registered.

■ Create a word sound pattern to be registered by concatenating monosyllabic patterns that have been registered in advance.

Alternatively, monosyllabic patterns are registered, and during manual word speech recognition, the word speech pattern created by concatenating the registered monosyllabic patterns is compared with the input speech pattern.

Note that each word is made up of syllables, and each syllable is usually formed by combining one vowel and one consonant, and there are about 100 types.

[Problem that the invention seeks to solve]

As a method for creating a word speech pattern to be registered in a conventional word speech recognition device in which the double type of the recognition target is not limited, the methods (1) and (2) described above have been used.

Method ■ has the advantage of high recognition rate, but on the other hand,
Since the user needs to register the word sound patterns of the word group to be recognized in advance, in the case of a double type with a large number of words, it takes a lot of time and effort to register, and it also takes a lot of time and effort to register. There was a problem in that the registration had to be re-registered every time there was a change in the etc.

On the other hand, method ■ requires only the registration of at most 100 monosyllable patterns, so the registration process is simple.
Moreover, - Registering a variable syllable pattern has the advantage that even if the number or category of double types to be recognized changes, there is no need to re-register the monosyllabic pattern in the double type, With word phonetic patterns created by concatenating patterns, it is extremely difficult to take into account deformations such as articulatory combinations and devoicing that appear in words.
There was a problem that the recognition rate was lower than in the case of .

In addition, in articulatory coupling, when voices are uttered continuously, the shape of the vocal tract cannot change suddenly, so frequency shifts occur due to the influence of the preceding and following phonemes (vowels or consonants), and the difference between phonemes and phonemes. This is a phenomenon in which the acoustic properties change continuously in the middle of the range. Also, devoicing is the vowel “i” and “U” sandwiched between voiceless consonants.
” is a phenomenon in which the vowel is devoiced, and this devoicing of vowels is an indispensable process for giving naturalness to synthesized Japanese sounds.

SUMMARY OF THE INVENTION An object of the present invention is to solve the problems in the conventional word speech recognition apparatus and provide a word speech recognition apparatus that can obtain a better recognition rate with a reduced amount of registration work.

[Means for solving problems]

The means taken by the present invention to solve the above-mentioned problems in conventional word speech recognition devices will be explained with reference to FIG.

FIG. 1 is a block diagram showing the basic configuration of the present invention.

In FIG. 1, 110 is a pseudo word pattern dictionary section;
Pseudo-word speech patterns created by connecting monosyllabic patterns are registered in advance.

Reference numeral 120 denotes a specific word pattern dictionary section in which a specific word sound pattern is registered in advance for a specific word, which is created based on the word sound actually uttered.

Reference numeral 130 denotes a word recognition unit, which inputs the input word speech pattern created from the input word speech to the pseudo word pattern dictionary unit 1.
10 and the specific word pattern dictionary section 120 to perform word recognition.

[For production]

Prior to recognition processing of input word sounds, pseudo word sound patterns created by concatenating monosyllabic patterns are registered in the pseudo word pattern dictionary section 110, as in the past.

On the other hand, in the word pattern dictionary section 120, specific word speech patterns created based on word sounds actually uttered for specific words are registered in advance. The specific words include words that have a large influence on effective recognition performance, such as words that appear frequently, important words, and words that are easily misunderstood. Since the number of these specific words is not large, even if the user actually utters the words and registers the word pattern, the amount of registration work is small.

The word recognition section 130 combines the input word speech pattern created from the input word speech to be recognized with the word speech pattern registered in at least one of the pseudo word pattern dictionary section 110 and the word pattern dictionary section 120. Verify and perform word recognition.

The input word voice pattern is input to both the dictionary units 110 and 12.
The recognition rate can be improved by matching the word sound patterns of both words registered as 0.

By doing the above, a better recognition rate can be obtained with less registration work.

〔Example〕

An embodiment of the present invention will be described with reference to FIG.

FIG. 2 is a block diagram showing the configuration of an embodiment of the present invention.

(A) Structure of the Embodiment In FIG. 2, the pseudo word pattern dictionary section 110, the specific word pattern dictionary section 120, and the word recognition section 130 are as described in FIG.

140 is a microphone into which words or monosyllables uttered by a speaker are input.

Reference numeral 150 denotes a parameter extraction unit, which extracts parameters representing the characteristics of word sounds or syllables input from the microphone 140.

Reference numeral 160 denotes a section detection unit that performs section detection processing on the input word sounds or syllables to create word patterns or monosyllable patterns.

170 is a monosyllabic pattern dictionary section in which monosyllabic patterns are registered in advance.

180 is a switching circuit which supplies the output from the section detection section 160 to the specific word pattern dictionary section 120 or the monosyllabic pattern dictionary section 170 during registration, and to the word recognition section 130 during word recognition.

(B) Operation of the Embodiment The operation of the embodiment will be explained separately into a pseudo word speech pattern registration operation, a specific word speech pattern registration operation, and a word recognition operation.

(B-1) Pseudo-word speech pattern registration operation During the pseudo-word pattern registration operation, the switching circuit 180 is connected to the monosyllable pattern dictionary section 170 side.

The microphone 140 inputs each monosyllable (approximately 100 types) uttered by a speaker (not shown) to the parameter extraction unit 50.

The parameter extraction unit 150 extracts parameters representing the characteristics of the input single syllable. Furthermore, the section detection unit 160 performs section detection based on the extracted parameters to create a monosyllable pattern. The configuration and operation of the parameter extracting section 150 and section detecting section 160 are well known, so a detailed explanation thereof will be omitted.

The created monosyllabic pattern passes through the switching circuit 180 and is registered in the monosyllabic pattern dictionary section 170.

Once a group of words to be recognized, for example the category of double electric, is determined, the monosyllabic patterns registered in the monosyllabic pattern dictionary section 170 are concatenated to create the speech pattern (pseudo word speech pattern) of each word belonging to that category. ) is automatically created and registered in the pseudo word pattern dictionary section 110.

By doing so, even if the pseudo word pattern dictionary section 110 is configured with a small memory capacity, it is possible to recognize word groups of various categories, so the word recognition device is not limited to the double voltage to be recognized. can be configured.

Note that the process of automatically creating a pattern for each word by concatenating monosyllabic patterns is well known, so a detailed explanation thereof will be omitted.

(B-2) Specific word speech pattern registration operation The specific word speech pattern registered in the word pattern dictionary section 120 in the specific word speech pattern registration operation has a large influence on the effective recognition performance among the word group to be recognized. This is a pattern of specific words, such as frequently appearing words, important words, and words that are easily misunderstood. These specific word sound patterns created based on the word sounds actually uttered by the speaker are registered.

During the specific word voice pattern registration operation, the switching circuit 180
is connected to the specific word pattern dictionary section 120 side.

The microphone 140 inputs each specific word voice uttered by the speaker to the parameter extraction unit 150 .

The parameter extraction unit 150 extracts parameters representing the characteristics of the input specific word speech. Furthermore, the section detection unit 160 performs section detection based on the extracted parameters to create a specific word speech pattern for the specific word. These processes are common to the monosyllable pattern creation process described above.

The created specific word voice pattern is transferred to the switching circuit 180.
The word pattern is registered in the specific word pattern dictionary section 120 through the word pattern dictionary section 120.

(B-3) Word recognition operation During word recognition operation, the switching circuit 180
Connected to the 0 side.

The microphone 140 inputs word sounds uttered by the speaker to the parameter extraction unit 150 .

The parameter extraction unit 150 extracts parameters representing the characteristics of the input word sound, and the interval detection unit 160 performs interval extraction to create an input word sound pattern. ! Recognition unit 13
Enter 0. These input word audio pattern creation processes are the same as the specific word audio pattern creation processing described above.

The word recognition unit 130 compares the input word voice pattern with each pseudo word voice pattern registered in the pseudo word pattern dictionary unit 110 and each specific word voice pattern registered in the word pattern dictionary unit 120, and Outputs the word with the smallest distance among the registered patterns as the recognition result.

When recognizing a specific word, use the specific word pattern dictionary section 1.
20 specific word sound patterns, and other words may be compared with each pseudo word sound pattern of the pseudo word pattern dictionary section 110, but as described above, both the dictionary sections 110 and 120 Better recognition results can be obtained by comparing registered word sound patterns together.

Although one embodiment of the present invention has been described above, the method of the present invention can also be applied to other pattern matching devices, such as character recognition and handwritten figure recognition devices.

For example, in the case of a character recognition device, standard stroke patterns are registered in advance, and once a group of characters to be recognized is determined, the registered standard stroke patterns are concatenated to create a pseudo character pattern and registered.

On the other hand, for characters that have a large impact on effective recognition performance, such as characters that appear frequently, important characters, and characters that are easily misunderstood, the specific character pattern of the actual character or actually written character is used. Create and register.

During character recognition, the input character pattern created for the input character is compared with each of the registered pseudo character patterns and specific character patterns, and the character with the smallest distance is output as the recognition result. .

As a result, as in the case of word speech recognition, a better recognition rate can be obtained with a smaller amount of registration work.

〔Effect of the invention〕

As explained above, according to the present invention, compared to conventional word speech recognition devices, it is possible to obtain a better recognition rate with a smaller amount of registration work.

[Brief explanation of drawings]

Fig. 1...Explanatory diagram of the basic configuration of the present invention, Fig. 2...
FIG. 3 is an explanatory diagram of the configuration of an embodiment of the present invention; FIG. 3 is an explanatory diagram of a conventional word speech recognition device. 1 and 2, 110...pseudo word pattern dictionary section, 120...specific word pattern dictionary section, 130...vehicle language recognition section, 14
0...Microphone, 150...Parameter extraction section,
160... section detection section, 170... monosyllabic pattern dictionary section. The story of the 10th pole of the 2nd picture of the Hanawazu fork of the Baoqiu Shu-? P. No. 3, Iron Home Village

Claims (2)

[Claims] (1) In a word speech recognition device that recognizes input word speech by comparing pre-registered word speech patterns with input word speech patterns created from input word speech, (a) Concatenates monosyllabic patterns. (b) a pseudo word pattern dictionary section (110) in which pseudo word sound patterns created based on the word sounds actually uttered are registered; A specific word pattern dictionary section (120) that has been registered in advance; A word speech recognition device comprising: a word recognition unit (130) that performs word recognition by comparing it with a word speech pattern registered in at least one of the words. (2) The word recognition section (130) converts the input word speech pattern created from the input word speech into the pseudo word pattern dictionary section (110) and the specific word pattern dictionary section (120).
2. The word speech recognition device according to claim 1, wherein the word recognition device performs word recognition by comparing both word speech patterns registered in .