JPH0210957B2 - - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention converts input speech into a plurality of candidate strings based on information obtained by a speech recognition system, particularly a speech analysis device, and converts input speech into a plurality of candidate strings using a word dictionary and at least a grammar connection matrix. The present invention relates to a speech recognition system that improves the recognition rate by indexing the candidate string, converting the candidate string into a meaningful word string, and extracting the word string as a grammar-checked word string.

Traditionally, speech recognition systems check syllables of input speech against a feature dictionary.
Categories with high similarity are output as recognition results.

However, it is generally difficult to unambiguously determine the category of each syllable, and correct recognition results may not always be obtained.

The purpose of the present invention is to solve the above-mentioned problems, and to improve the recognition rate by extracting a plurality of candidate sequences, for example, a word sequence that is a string of words that have meaning in Japanese. The purpose is to improve. Therefore, the speech recognition system of the present invention includes a speech analysis device that analyzes the speech input signal based on the speech input signal and recognizes the input speech. The system is configured to extract a plurality of candidate syllables for each syllable in the order of pronunciation, and to add similarity information corresponding to each candidate syllable, and to select one of the candidate syllables for each syllable. a candidate string creation device that creates candidate strings in the order of selecting items with high similarity based on the similarity information; a candidate string analysis device; and a word string extracted by the candidate string analysis device. The word string buffer is configured to have a word string buffer that is assigned a priority order and stored, and an output display device control unit that outputs the contents of the word string buffer. indexing a word dictionary having at least grammatical information, frequency information, and word information;
The system is configured to convert the input candidate string into a meaningful word string while indexing the grammatical connection matrix for the extracted words based on the grammatical information, and converts the input speech into a word string. and the output display device control section is configured to output the word string buffer and give a priority to each word string, and the output display device control section is configured to display the contents of the word string buffer in the order of the priority. It is a feature. This will be explained below with reference to the drawings.

FIG. 1 is an overall configuration diagram of an embodiment of the present invention, and FIG. 2 is a configuration diagram of an embodiment of the candidate sequence creation device shown in FIG. 1.
FIG. 3 shows the configuration of an embodiment of the candidate sequence analysis device shown in FIG.

In FIG. 1, 1 is a microphone, 2 is a speech analysis device which has a conventionally known configuration, for example, to perform frequency analysis, extract feature parameters, and check them with a feature dictionary; and 3, a speech analysis device 2. A syllable candidate buffer stores, for example, candidate monosyllables for each monosyllable that have been analyzed, and 4 is a candidate string creation device that stores candidate monosyllables stored in the syllable candidate buffer 3 for each monosyllable of input speech. a candidate string creation device that selects one and creates a candidate string; 5 is a candidate string buffer; 6 is a candidate string analysis device that extracts a meaningful word string based on the contents of the candidate string buffer; 7 8 is a word string buffer for storing extracted word strings, 8 is an output display device control unit, and 9 is a display.

As the speech analysis device 2 is conventionally known, a detailed explanation will be omitted, but assuming that the input speech is composed of, for example, seven syllables, the speech analysis device 2 analyzes candidate word phrases for each syllable and generates similarity information. It is made to be attached and extracted. In the case shown, for the first syllable,
The candidate monosyllables "te", "ke", "he", and "e" are extracted, and the similarity information for each candidate monosyllable is extracted as "1", "2", "5", and "9". has been done.
Regarding the second syllable, candidate monosyllables "i", "pi",
“K” and “H” are extracted and their similarity information is “0”,
"3", "4", and "5" are extracted.

The candidate string creation device 4, which will be described in detail with reference to FIG. 2, creates combinations by selecting one candidate monosyllable for each syllable of the input speech, and selects combinations with the smallest total value of the similarity information. Candidate sequences ``Teyausarel'', ``Keyausarel'', ``Teyapsarel'', ``Teiyausorel'', etc. are created by arranging them in order.

The candidate string analysis device 6, which will be described in detail with reference to FIG. In this case, priority is assigned to the obtained word strings based on the extraction order of the candidate strings and frequency information about words in a word dictionary, which will be described later.

The output display device control section 8 displays the word string on the display 9 in the form of a character pattern. At this time, the images are displayed in order according to the priority order.

FIG. 2 shows the configuration of an embodiment of the candidate sequence creation device shown in FIG. Reference numeral 5 in the figure corresponds to FIG.
10-1 to 10-7 are candidate monosyllable groups, 1
Reference numerals 1-1 to 11-7 represent monosyllable reading sections, 12 a candidate string generation register, and 13 a similarity addition section. Monosyllable reading section 11-1 to 11-
7 reads out one from each candidate monosyllable group and supplies it to the candidate string generation register 12. At this time, for example, the first monosyllable from each group 10-1 to 10-7 is read out to extract "teiyausarel", then only group 10-1 is taken as the second monosyllable to extract "keyausalel", and the next monosyllable is read out. Then, change only group 10-1 to the first monosyllable, change only group 10-2 to the second monosyllable, take out "tepiyausarel", and so on,
The combinations that are considered to have the smallest total similarity value are taken out first, so to speak, all the combinations are created one after another. Candidate string generation register 12
Similarity information is added to the candidate strings generated by the similarity adding section 13, and candidate strings having smaller values are sequentially stored in the candidate string buffer 5.

FIG. 3 shows the configuration of an embodiment of the candidate sequence analysis device shown in FIG. Reference numerals 6 and 7 in the figure correspond to those in FIG. 1, G is a gate, 14 is an input register, 15 is an analysis processing unit, 16 is a word dictionary, 17 is a grammar connection matrix, 18 is a semantic connection matrix, and 19 is a dictionary In the search device, 20 represents a connection check processing section, and 21 represents a word string check section.

One candidate string from the candidate string buffer 5 shown in FIG. 1 is set in the input register 14 and processed by the analysis processing section 15. In the case shown in the figure, since the candidate string is "te", "i", "ya", etc. in order from the beginning, the word dictionary 16 is indexed via the dictionary search device 19, and the word is searched using the longest match method. Words are sequentially read out from the word dictionary 16 based on a maximum evaluation method that combines the frequency of occurrence of the words.

Note that the word dictionary 16 has readings 22 and word information 2.
3. Grammar information 24, semantic information 25, frequency information 26
etc. are stored as one word.

In the illustrated case, words such as "body,""bottom," and "hand" are read out, and grammatical information about each word is read out together. Next is “body” and “bottom”
The word ``ya'' is read out in response to ``te,'' and ``ya'' and ``i'' are read out in response to ``te.'' Then, a grammatical connection check and a semantic connection check are performed via the connection check processing section 20.

What is illustrated in the analysis processing unit 15 shown in FIG. 3 represents the form of connection between words.
- "Ya" indicates that connection is possible, "Hand"
*“No” indicates that connection is not possible.

Note that the grammar connection matrix 17 can be considered to be composed of a matrix in which grammar information is arranged on the vertical and horizontal axes, and in the case shown in the figure, for example,
(i) Words with grammatical information ``1'' are not grammatically connected to words with grammatical information ``53'' (logical ``0''), and (ii) words with grammatical information ``3'' are grammatically connected with words with grammatical information ``53''. ” (logic “1”), and (iii)
A word with grammatical information ``53'' can be connected to a word with grammatical information ``44,'' indicating that... For this reason, for example, from the dictionary 16, (i) grammar information "3"
If the word ``adjective'' with ``adjective'' is read out, (ii) then the word ``are'' with grammatical information ``53'' is read out, and (iii) then the word ``ru'' with grammatical information ``44'' is read out. In the result of indexing the grammar connection matrix 17, "adjective" is obtained as one grammatically connectable word. Note that an aspect of grammar checking using the grammar connection matrix 17 is disclosed in Japanese Patent Application No. 1982-92270.

In addition, the semantic connection matrix 18 can be thought of as a matrix in which semantic information is arranged on the vertical and horizontal axes in the same manner as the grammatical connection matrix 17, and the intersection points are arranged in correspondence between words that can be semantically connected. Logic "1" is written in Regarding the semantic check using the semantic connection matrix, please refer to the patent application 1986-
It is disclosed in No. 102636, and a detailed explanation will be omitted.

As mentioned above, the analysis processing section 15 shown in FIG.
Index the word dictionary 16 and create the grammar connection matrix 17
and the semantic connection matrix 18, and processes are performed to convert the input candidate strings into meaningful word strings. When the processing for one candidate string is completed, the gate G is controlled to set the next candidate string in the input register 14, and the obtained meaningful word string is stored in the word string buffer 7.
At this time, the obtained word string is added according to its priority order. The configuration of the analysis processing section 15 is disclosed in Japanese Patent Application No. 1983-92271.

When candidate monosyllable groups 10-1 to 10-7 as shown in FIG. 2 are obtained, as a result,
Words such as "described", "used in combination", etc. will be set in the word string buffer 7.

As explained above, according to the present invention, the analysis result by the speech analysis device can be checked grammatically and semantically if necessary, and extracted as a meaningful word string. The recognition rate is greatly improved compared to the recognition rate obtained only by using this method. In addition, the candidate string creation device and candidate string analysis device each output information in the order of similarity and priority, which reduces the possibility that necessary candidates overflow from the word string buffer, resulting in more efficient processing. It can be performed.

[Brief explanation of drawings]

FIG. 1 is an overall configuration diagram of an embodiment of the present invention, and FIG. 2 is a configuration diagram of an embodiment of the candidate sequence creation device shown in FIG. 1.
FIG. 3 shows the configuration of an embodiment of the candidate sequence analysis device shown in FIG. In the figure, 2 is a speech analysis device, 4 is a candidate string creation device, 6 is a candidate string analysis device, 7 is a word string buffer,
15 is an analysis processing unit, 16 is a word dictionary, 17 is a grammar connection matrix, 18 is a semantic connection matrix, 1
Reference numeral 9 represents a dictionary search device, and reference numeral 20 represents a connection check processing section.

Claims (1)

[Scope of Claims] 1. A speech recognition system having a speech analysis device that analyzes the speech input signal based on the speech input signal and recognizes the input speech, wherein the speech analysis device utters the input speech. It is configured to sequentially extract a plurality of candidate syllables for each syllable and provide similarity information corresponding to each candidate syllable, and select one of the candidate syllables for each syllable to provide the similarity information. a candidate string creation device that creates candidate strings that are combined in the order of selection based on the highest similarity; a candidate string analysis device; and a candidate string analysis device that gives priority to the word strings extracted by the candidate string analysis device. The candidate string analysis device is configured to include a word string buffer for storing the word string buffer, and an output display device control unit for outputting the contents of the word string buffer. indexing a word dictionary having frequency information and word information;
The system is configured to convert the input candidate string into a meaningful word string while indexing the grammatical connection matrix for the extracted words based on the grammatical information, and converts the input speech into a word string. and the output display device control section is configured to output the word string buffer and give a priority to each word string, and the output display device control section is configured to display the contents of the word string buffer in the order of the priority. Features a voice recognition system.