JPH0656559B2 - Word detection method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention is used in a voice recognition device, a voice input device, etc., and is a word for detecting a word included in an input voice and a position of the word in the voice. Regarding detection method.

(Prior Art) A method for detecting a word and its position in an input voice in a voice recognition device, a voice input device, etc., includes a method for inputting each category from the input voice that is a sequence of categories such as syllable, phoneme, and phoneme class Position information in the voice is extracted, and if the category string created from the extracted categories corresponds to the category string of a word, that word and the position of the category string in the input voice are output as the detection result. There is a way.

In general, the above-mentioned categories have a short time length, and there are similar categories. Therefore, it is difficult to extract the categories in the input speech completely without error.
Therefore, conventionally, a plurality of category candidates are extracted for each category section in the input voice, and the category candidates are sequentially used from the end of the input voice to generate a partial category candidate sequence. Found a category candidate sequence corresponding to the word by repeating the process of matching the word with the category sequence. For details of this method, refer to, for example, Document 1 “Japanese Patent Application No. 58-214544, Pattern Recognition Device”.
Since it has been described above, it is omitted here.

In addition, at the stage of extracting the categories in the input voice, a category that is not included in the input voice is detected due to factors such as a slur of the utterance and articulation of adjacent categories (for example, syllables). Other category candidates may appear between the power category candidates. This phenomenon is hereinafter referred to as category insertion error.

Conventionally, the following method has been used to deal with a category insertion error. That is, it is preliminarily investigated what kind of category arrangement is likely to cause a category insertion error, and from the result, a category string correction rule is prepared for a relatively frequent category insertion error. When the category string correction rule is applied to the category string in which the category insertion error has occurred, the inserted category is deleted. By applying this rule to the category candidate sequence at the time of word detection, it is possible to correct category insertion errors that are relatively frequent. This method is described in, for example, Reference 2 “Shoichi Matsunaga, Masanori Yoshida“ Candidate Selection Using the Effects of Branch & Bound and Bottom-Up Phonological Recognition ”, Acoustical Society of Japan Material S85-79, January 1986”. Pp. 616, lines 15-18. An example of the correction rule is shown in Table 4 on page 617 of the same document.

(Problems to be Solved by the Invention) In the above-mentioned conventional method, since the category candidate string is generated using the category candidates extracted from the input speech, the category candidate string is collated with the word category string. A large number of category candidate sequences, which are wasted, are generated, which requires a large amount of calculation.

Further, even if a word segment to be detected occupies only a part of the input voice, conventionally, all the category candidates including the segment in which the word does not exist are matched equally with the category in the word from the end of the input voice. Must be done,
It takes a lot of calculation time and a long time to detect a word.

Furthermore, the above-mentioned category string correction rule is equally applied not only to the category candidate string in which the category insertion error has occurred but also to the category candidate string in which the category insertion error has not occurred. Further, in many cases, a plurality of correction rules are individually applied to one category candidate sequence. Therefore, many category candidate strings are generated from one category candidate string, and the number of category candidate strings to be matched with the word category string increases. Moreover, most of the category candidate strings are rejected because they do not match the category string of the word to be detected.

Moreover, category insertion errors that can be corrected by the correction rule are limited to those that occur relatively frequently, and errors that occur infrequently cannot be corrected. The number of correction rules must be increased in order to increase the types of errors that can be corrected, but as a result, more and more category candidate sequences are generated.

For example, when a voice pronounced "Onsei Ninshikiwa (voice recognition is used)" is input and syllable candidates are extracted, the syllable "s" is inserted between the syllables "n" and "si". D ”may be inserted by mistake. In this case, even if the correct syllable candidates are obtained for other syllables, the syllable candidate sequence generated from the extracted syllable candidates becomes “Onsei Ninishikiwa”, and the correct word candidate “recognition The word “recognition” cannot be detected because there is no portion that matches the syllable example “ninshinki”. Moreover,
Such a syllable insertion error is a relatively rare kind, and there are few rules for correcting this error.

An object of the present invention is not to generate a wasteful category candidate sequence, and also when a section of a word to be detected is a very small part of the entire input speech, or when some category is detected when a category candidate is detected in the input speech. Even if the candidate was accidentally inserted,
It is an object of the present invention to provide a word detection method that enables efficient detection of a correct word and its position from input speech.

(Means for Solving Problems) Means provided by the present invention for solving the above problems and achieving the above object are extracted from an input speech that is a sequence of categories such as syllables, phonemes, and phoneme classes. A word detection method for detecting a word in an input voice and its appearance position by generating a category candidate string corresponding to a word category string using a plurality of category candidates and their position information, A plurality of category candidates obtained from the input voice and their position information are classified and stored by the category name, and the category candidates and their position information corresponding to each category are stored according to the order of the categories in the word. While selecting from the category candidates that have been classified into the same name as the category and stored, each of the two adjacent categories in the word has three consecutive words in the input speech. When it corresponds to the category candidates at both ends of the arrangement of the individual category candidates, the arrangement of the two categories and the arrangement of the three category candidates are made to correspond to each other to generate the category candidate sequence.

(Operation) In the method of the present invention, among the category candidates extracted from the input speech, only the category candidates having the same name as the category included in the word to be detected are used, and the arrangement of the categories in the word is followed. A category candidate sequence to be generated is generated. As a result, only the category candidate string corresponding to the word category string or the substring thereof is generated, and it is possible to avoid generating an unnecessary category string.

In addition, in order to generate a category candidate string from category candidates corresponding to the category in a word among the category candidates in the input speech, when the section of the word to be detected is a very small part of the entire input speech. It is possible to detect the word quickly regardless of whether or not the section exists in the input voice.

Further, even if a category insertion error occurs, it is possible to detect a word according to the following principle. Now, as a result of extracting the category candidates in the input speech including the word W whose category string is C ₁ ... C _i-1 C _i ... C ₁ , the categories C _i-1 ,
It is assumed that the category candidate K _X is erroneously inserted between the category candidates K _i-1 and K _i of C ₁ . That is, the category candidate sequence of the portion corresponding to the word W in the input voice is K ₁ ... K _i-1 K _X K _i ... K ₁ . Therefore, when generating a category candidate string corresponding to the word while following a sequence of categories in the word to be detected, corresponding to the category candidate (K _i-1) and category C ₁ corresponding to the category C _i-1 If the category candidate (K ₁ ) to be processed is a category candidate at both ends of the sequence of three consecutive category candidates in the input speech, the sequence of these three category candidates is the category string C _i in the word. _-1 Corresponds to C ₁ . This also enables correct correspondence between the category candidate sequence and the category sequence even if the category candidate K _X is inserted. Further, since only the category candidate string corresponding to the word category string is generated, useless generation of the category candidate string can be avoided.

(Examples) Hereinafter, the present invention will be described in more detail according to examples with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention.

In this embodiment, Japanese voice is input and syllables are used as categories. The syllable extraction unit 101 detects a syllable candidate in the input voice and stores the candidate in the syllable candidate storage
Store in 102.

An example of the syllable extraction unit 101 is shown in a block diagram in FIG. In FIG. 2, the input voice is temporarily stored in the voice buffer 201. First, the vowel candidate detection unit 202 uses the voice buffer 201.
The vowel candidate in the voice stored in is detected and stored in the vowel candidate storage unit 203. Vowel candidate detection is performed by the vowel pattern storage unit 2
This is performed by comparing the standard voice pattern of each vowel stored in 04 with each section of the input voice. The vowel voice signal is relatively stationary and therefore easy to detect. Each vowel candidate holds at least vowel name and position information in the input voice. After the detection of vowel candidates is completed, the consonant candidate detection unit 205 detects consonant candidates as described below. In Japanese, a syllable is a consonant (C) -vowel (V) pair. Therefore, in the input speech, a section between two vowels that is shorter than a certain time length (this is a VCV section) and a section from the beginning of the input speech to a vowel that is within a certain time length (this is a CV section). It can be said that there is one consonant in each (section). The consonant candidate detection unit 204 preliminarily stores the consonant pattern storage unit 20 for all VCV sections and CV sections created from the vowel candidates stored in the vowel candidate storage unit 203.
The VCV and CV standard voice patterns stored in 6 are collated, and the names of a plurality of voice patterns having a high degree of similarity are used as consonant candidates. The vowel candidate and the consonant candidate determined as described above are combined into a syllable candidate and stored in the syllable candidate storage unit 102 together with the position in the input voice.

As an example, if the voice "ON-SEI NINISHIWA" (voice recognition) is input, syllable candidates as shown in FIG. 3 are extracted as a result of the syllable recognition. In FIG. 3, a line of an arrow is a section of each syllable candidate, and a plurality of syllable candidates are extracted in each section. These syllable candidates are classified by syllable name, and the syllable candidate storage unit 102
Remembered. As a result, the contents of the syllable candidate storage unit 102 are as shown in FIG. In this figure, each syllable candidate is represented in the format of "syllable name / start time: end time".

The word storage unit 103 stores syllable strings of words to be detected. After fetching one of the words into the word buffer 104, it is examined whether or not this word is included in the input speech. Now, it is assumed that the word buffer 104 stores the syllable string “Ninshiki” of the word “recognition”.

The syllable candidate string generation unit 105 creates a syllable candidate string from the syllable candidates in the syllable candidate storage unit 102 in the order of the arrangement of syllables in the words stored in the word buffer 104, and corresponds to the resulting syllable candidate string. The syllable sequence and the syllable candidate sequence storage unit 106 are stored. In this embodiment, a syllable string is created in order from the beginning syllable of a word.

First, since the syllable at the beginning of the word buffer 104 is “d”, the syllable candidate string generation unit 105 extracts the syllable candidates stored in the syllable candidate storage unit 102 classified into “d” and stores the length of each syllable candidate. As the syllable candidate sequence of 1, the syllable "D"
It is stored together with the syllable candidate string storage unit 106. As a result,
The syllable candidate string storage unit 106 stores four syllable candidate strings: 0/2 (d) 2/2: 4 (d) d / 10: 12 (d) d / 14: 16 (d). To be done. Here, the brackets are the corresponding syllable strings.

Next, the syllable candidate string generation unit 105 pays attention to the next syllable “n” in the word buffer 104, and for each of the syllable candidates stored in the syllable candidate storage unit 102 classified as “n”, Immediately after the last syllable candidate in any of the syllable candidate strings in the candidate string storage unit 106, or via another one syllable candidate, it is checked whether or not the input speech is followed. If there is a syllable candidate that continues like that,
The syllable candidate string is connected to the tail end of the syllable candidate string to generate a new syllable candidate string and stored in the syllable candidate string storage unit 106. Whether or not the syllable candidate A follows the other syllable candidate B can be determined by comparing the end time of the syllable candidate A and the start time of the syllable candidate B. Here, when the difference between those times is plus or minus 1 or less, it is determined to follow. In the present case, there are three syllable candidates that are classified and stored as "n", that is, n / 2: 4 and n / 12: 14. Therefore, since the syllable candidate / 2: 4 follows immediately after the last syllable candidate d / 2: 0: 2 of the syllable candidate string, the syllable candidate string is connected to the syllable candidate d / 2: 0: 2. Nin ”. Similarly, the syllable candidate / 12: 14 is connected to the syllable candidate string to form the syllable string “nin”. In addition, the syllable candidate string stored in the syllable candidate string storage unit 106 until then is deleted. As a result, in the syllable candidate string storage unit 106, there are two syllable candidate strings of N / 0: 2-N / 2: 4 (Nin) N / 10 / 10-12-N / 12: 14 (Nin). Remain.

Then, the process proceeds to the syllable "si". In the syllable candidate storage unit 102, there are two syllable candidates classified and stored as “si”, that is, si / 4: 7 and si / 16: 18. For each of these, it is checked whether the last syllable candidate with the syllable candidate sequence is succeeding in the input voice directly or via another one syllable candidate. As a result, Si / 4: 7 is connected to the syllable candidate sequence. See also / 16:18
Is connected to the last syllable candidate / 12: 14 of the syllable candidate sequence via syllable candidate di / 14: 16 or a / 14: 16, so syllable candidate sequence / 16: 18 becomes the syllable candidate sequence. Be connected.
Therefore, the contents of the syllable candidate string storage unit 106 are: Ni / O: 2-N / 2: 4-Si / 4: 7 (Ninshi) Ni / 10: 12-N / 12: 14-Si / 16: 18 (Ninshi) ).

Here, the process proceeds to the last syllable “ki” in the word buffer 104. In the syllable candidate storage unit 102, the syllable candidates that are classified and stored as “ki” are ki / 18: 19 1
It is an individual. For this syllable candidate, it is checked whether the syllabic candidate at the end of the syllable candidate sequence is followed in the input speech directly or via one other syllable candidate. In this case, the last syllable candidate sequence in the syllable candidate sequence /
It follows immediately after 14:16. Therefore, the syllable candidate key / 18: 19 is connected to the syllable candidate string, and the next syllable candidate string is stored in the syllable candidate string.

D / 10: 12- / 12/12: 14-Si / 16: 18-key / 18: 19
(Ninshiki) Here, since the last syllable in the word buffer 104 has been reached, the syllable candidate string generation unit 105 outputs that the word “recognition” exists in the section from time 10 to time 19 in the input speech. To do.

The embodiment of the present invention has been described above. A plurality of category insertion errors may occur in one word unless they are consecutive.

(Effects of the Invention) As described above, according to the present invention, even when some extra syllable candidates are erroneously inserted at the stage of syllable candidate extraction from input speech, the existence and input of the word are input. It becomes possible to detect the position in the voice, and the number of syllable candidate strings generated during the detection process is extremely small,
A word detection method that enables efficient word detection can be provided.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment of the present invention, FIG. 2 is a block diagram showing a concrete example of a syllable extraction unit in the FIG. 1 embodiment, and FIG. 3 is an input voice and extraction in the FIG. 1 embodiment. FIG. 4 is a diagram showing an example of the selected syllable candidates, and FIG. 4 is a diagram showing an example of the contents of the syllable candidate storage unit in the embodiment of FIG. 101 ... syllable extraction unit, 102 ... syllable candidate storage unit, 103 ...
Word storage unit, 104 ... word buffer, 105 ... syllable string generation unit, 106 ... syllable string storage unit, 201 ... voice buffer, 202
Vowel candidate detection unit, 203 vowel candidate storage unit, 204
Vowel pattern storage unit, 205 ... Consonant candidate detection unit, 206 ... Consonant pattern storage unit.

Claims (1)

[Claims] 1. A category candidate sequence corresponding to a word category sequence is generated using a plurality of category candidates extracted from an input speech, which is a sequence of categories such as syllables, phonemes, and phoneme classes, and their position information. By doing so, in a word detection method for detecting a word in an input voice and its appearance position, each of a plurality of category candidates and their position information obtained from the input voice is classified and stored by the category name, and the word is stored. The category candidates corresponding to each category and their position information are selected from the category candidates that are classified and stored under the same name as the category according to the order of the categories in the list, and two adjacent categories in the word are selected. , Respectively correspond to the category candidates at both ends of the sequence of three consecutive category candidates in the input voice, the sequence of the two categories Three of the sequence of category candidates to correspond, word detection system, characterized in that for generating the category candidate string.