JPS63265299A - 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 [Field of Industrial Application] The present invention relates to a speech recognition device that sequentially recognizes speech information syllable by syllable.

[Prior Art] Conventionally, this type of speech recognition device has a syllable dividing means for dividing input speech information into syllables, a vowel recognition means for recognizing the vowel parts of the divided syllables, and various standard consonant patterns. and consonant recognition means for recognizing the consonant part of the divided syllable, and the consonant recognition means stores the characteristic pattern of the consonant part of the divided syllable and the consonant part of the divided syllable. The consonant part is recognized by matching with a standard consonant pattern.

[Problems to be Solved by the Invention] In the above-mentioned consonant recognition means, since it is necessary to match the characteristic patterns of the consonant parts of the classified syllables with all consonant standard patterns, the amount of processing required for consonant recognition is And the amount of time was enormous.

[Object of the Invention] The present invention has been made to solve the problem of (e) above, and is to reduce the number of matching operations in the consonant m'l 'I&, and to reduce the amount of processing and time required for recognition. The aim is to reduce

[Means for Solving the Problems] As shown in FIG. 1, the speech recognition device of the present invention includes a syllable dividing means 1 for dividing input speech information into syllables, and a vowel part of the divided syllables. A vowel recognition means 2, a consonant standard pattern storage means 6 that stores various standard consonant patterns, and a consonant standard pattern storage means 6 that stores various standard consonant patterns, match the characteristic pattern of the consonant part of the divided syllable with the consonant standard pattern to determine the syllable. and a consonant recognition means 5 which recognizes the consonant parts of the vowel recognition means 2 and the consonant recognition means 5. In order to achieve the above-mentioned purpose, the recognition results of the syllables recognized by the vowel recognition means 2 and the consonant recognition means 5 are chronologically analyzed. a syllable string creating means 7 which sequentially stores information on various words or phrases, a dictionary means 4 which stores information regarding various words or phrases, and a chronological sequence of the recognition results of the syllables stored in the syllable string creating means 7 and finally recognition. Consonant candidate creation, which compares the combination with the recognition result of the vowel part of the next syllable following the syllable with the information stored in the dictionary means 4, and creates a consonant candidate for the next syllable based on the verification result. The consonant recognition means 5 is configured to match only the characteristic pattern of the consonant part of the next syllable with the standard pattern of the consonant candidate among the various standard consonant patterns. It is a feature.

"Operation" The vowel recognition means 2 recognizes the vowel part of the syllable divided by the syllable division means 1.The consonant candidate creation means 3 recognizes the vowel part of the syllable divided by the syllable division means 1. Then, consonants that may form words or phrases stored in the dictionary means 4 are created as consonant candidates for the classified syllables.The consonant recognition means 5 uses only the consonant candidates as matching targets. The syllable string creating means 7 recognizes the consonant part of the divided syllable part as the vowel recognition means 2.
A new time series is constructed and stored from the time series of the recognition results of the syllables recognized by the consonant le recognition means 5 and the already stored time series of the recognition results of the syllables.

[Example] Examples of the present invention will be described below with reference to FIGS. 2 to 5.

The speech recognition device of this embodiment sequentially recognizes speech information input word by word syllable by syllable, and as shown in FIG. (Read-only memory) 20.
It consists of a RAM (random access memory) 30 and input/output capos 1 to 40. ROM20 is
A program 21 in which processing operations of the CPU 10 are described
．． It stores an EI phonetic phonetic pattern 22, a consonant standard pattern 23, and a word dictionary 24 in which various words are described in phoneme series, and the RAM 30 stores a phonetic information area 31.
, syllable interval pointer 32. An ffi sound register 33, a consonant candidate area 34, and a syllable string area 35 are formed. The syllable string area 35 is composed of three rows, as shown in FIG. 5, and each row stores a syllable string and its overall similarity. In addition, as shown in the same figure, the consonant candidate area 34 is composed of three rows like the syllable string area 35, and each row contains a syllable string based on the syllable string stored in the corresponding row of the syllable string area 35. The obtained consonant candidates and their degrees of similarity are stored. In this embodiment, a collection of city names is used as the word dictionary 24, and as shown in FIG. 4, the word dictionary 24 has a tree structure in which phonemes are nodes and connections between them are represented by branches. Furthermore, to make it easier to search for consonant candidates from the vowel recognition results, vowels and consonants are written in the reverse order from the normal Roman alphabet.

Next, the processing operation of the CPU 10 in the above configuration will be explained with reference to flowchart 1 in FIG.

Now, consider the case where "hamada" J is given as a word sound.The sound information input to the input/output port 40 is stored in the sound information area 31 in step S1.In step S2, the sound information is One syllable /ha/ is extracted, and the address of the audio information area 31 in which audio information corresponding to the beginning and end of the syllable is stored is stored in the syllable section pointer 32.Syllable section pointer 32
Regarding the first syllable designated by , the characteristic pattern of the vowel part /a/ is matched with the vowel standard pattern 22 in step S3, and the recognition result /a/ is shown in FIG. 5(a). It is stored in the vowel register 33. When recognizing the first syllable, since no syllable string is stored in the syllable string area 35, in step S4, the consonants /h/, / connected to the vowel /a/ of the first syllable are selected from the simple MFj word m24.
ni/, /m/, in>, and /S/ are stored in the first row of the consonant candidate area 34 as consonant candidates. In step S5, for the consonant candidates stored in the consonant candidate area 34, matching is performed between the characteristic pattern of the consonant part of the first syllable indicated by the syllable section pointer 32 and the standard pattern of the consonant candidate in the consonant standard pattern 23. is performed, and the degree of similarity determined thereby is stored in the consonant candidate area 34 following each consonant candidate [see FIG. 5(a)]. In step S6, the top three consonant IU complements stored in the consonant candidate area 34 are stored in each row of the syllable string area 35 as a syllable string along with the vowels stored in the vowel register 33 in descending order of similarity. At the same time, the similarity of each consonant candidate is stored as a total similarity [see FIG. 5(b)]. In step S7, it is determined whether it is the last syllable or not, and the process returns to step S2. In step S2, the second
The syllable /ma/ is extracted, the vowel part of the second syllable is identified in step S3, and the vowel /a/ is stored in the vowel register 33.See FIG. 1M5 (b). In step S4, first of all the syllable strings stored in the syllable string area 35,
Regarding the row /h a/, consonants /m/ and /n/ connected to the vowel /a/ of the second syllable are stored in the first row of the consonant candidate area 34 as consonant candidates of the second syllable, and Third
Consonant candidates are similarly stored in the consonant candidate area 34 for the syllable string of the row. In step S5, matching of the consonant part of the second syllable is performed for each consonant candidate so that there is no overlap, and the degree of similarity is stored in the consonant candidate area 34.
See FIG. 5(b)]. In step S6, for the combination of each syllable string and each consonant candidate found based on that syllable string, the sum of the overall similarity of the syllable string and the similarity of the consonant candidates is calculated, and the sum is determined in descending order of the sum. The first three digits are stored together with the vowel as a new syllable string in each row of the syllable string area 35, and at the same time, the sum is stored as the overall similarity of the new syllable string [see FIG. 5(C)]. From step S7, the process returns to step S2, and for the third syllable /CJa/, from step S2 to step S63. The processing is performed [see FIG. 5(C)]. If it is determined in step S7 that the syllable is the shortest, then in step S8 the & part sequence /hamada/, / stored in the syllable sequence area 35 is determined.
hamama/.

/5akata/ [see Figure 5 (, d)] which are complete words /hamada/, /5akat
Syllable string with the highest overall similarity among a//hamada
/ is output from the input/output port 40 as a recognition result.

[Effects of the Invention] As described in detail above, the speech recognition device of the present invention can recognize words or phrases stored in the dictionary means based on the recognition results of already recognized syllables and the recognition results of vowel parts. By matching only consonants that are likely to be formed, there is no need to perform unnecessary matching, and there is an advantage that the processing equipment and time required for consonant recognition can be reduced. Furthermore, according to the present invention, it is possible to recognize each syllable each time the syllable is input, so that speech information that has been input continuously in sentences without dividing it into words in advance can be recognized based on the recognition results for each syllable. This has the potential to be applied to a device that recognizes sentences in real time while analyzing their structure.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of a speech recognition device according to the present invention, FIG. 2 is a block diagram showing an embodiment of the present invention, FIG. 3 is a flowchart showing the processing operation of the CPU in this embodiment, and FIG. The figure shows a part of the memory contents of the word dictionary used in this embodiment, and Figure 5 shows changes in the memory contents of the vowel register, consonant candidate area, and syllable string area when word recognition is performed in this embodiment. FIG. syllable dividing means, 2: vowel recognition means, 3: consonant candidate creation means, 4: dictionary means, 5: consonant recognition means, 6: consonant standard pattern storage means, 7: @part sequence creation means.

Claims (1)

[Scope of Claims] 1. Syllable dividing means (1) for dividing input speech information into syllables; Vowel recognition means (2) for recognizing vowel parts of the divided syllables.
and a consonant standard pattern storage means (6) that stores various standard consonant patterns; and a consonant standard pattern storage means (6) that stores various standard consonant patterns; A speech recognition device comprising a consonant recognition means (5) for sequentially recognizing input speech information syllable by syllable, the speech recognition device comprising the vowel recognition means (2) and the consonant recognition means (5). syllable string creation means (7) for sequentially storing the recognition results of syllables recognized by the syllable strings in chronological order; dictionary means (4) for storing information regarding various words or phrases; and the syllable strings. A combination of the time series of syllable recognition results stored in the creation means (7) and the recognition results of the vowel part of the next syllable following the last recognized syllable is stored in the dictionary means (4). and a consonant candidate creation means (5) for creating a consonant candidate for the next syllable based on the verification result, and the consonant recognition means (5) is configured to identify the characteristic pattern of the consonant part of the next syllable. and a standard pattern of the consonant candidate among the various standard consonant patterns. 2. The consonant recognition means (5) determines the similarity of the standard pattern of each consonant candidate to the characteristic pattern of the consonant part of the next syllable by matching, outputs the similarity as a recognition result, and creates the syllable string. Means (7) combines each time series of already memorized syllable recognition results with the recognition result of the vowel part of the next syllable, and each consonant candidate obtained based on the combination. Construct a new time series from 2. The speech recognition device according to claim 1, wherein said new time series and their total similarity are stored up to a predetermined rank in descending order of similarity.