Speech recognition apparatus using syntactic and semantic analysis

• US 5,457,768 A
• Filed: 08/12/1992
• Issued: 10/10/1995
• Est. Priority Date: 08/13/1991
• Status: Expired due to Term

• Current Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Original Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Inventors: Tsuboi, Hiroyuki, Takebayashi, Yoichi
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): Dorvil, Richemond
• Application Number:
  US07/928,417
• Time in Patent Office: 1,154 Days
• Field of Search: 395/2.28, 395/2.12, 395/2.57, 395/2.6, 395/2.75, 395/2.4, 395/2.52, 381/42, 381/43
• US Class Current: 704/231
• CPC Class Codes: G10L 15/18 using natural language modelling
• CPC Class Codes: G10L 15/1815 Semantic context, e.g. disambiguation of the recognition hypotheses based on word meaning
• CPC Class Codes: G10L 15/19 Grammatical context, e.g. disambiguation of the recognition hypotheses based on word sequence rules
• CPC Class Codes: G10L 2015/088 Word spotting

Abstract

A speech recognition apparatus comprises a speech input unit for receiving an input speech signal, analyzing it, and outputting a speech feature parameter series, a speech recognition unit for extracting a speech feature vector from the parameter series, and matching it with a plurality of predetermined words to output a series of word candidates used as keywords, a syntactic analysis unit for analyzing the series of the word candidates as the keywords according to syntactic limitation, and generating a sentence candidate.

Claims

1. A speech recognition apparatus comprising:

speech analyzing means for receiving an external speech signal, analyzing the speech signal, and generating a speech feature parameter series;

dictionary means being storing a plurality of predetermined words having reference vectors;

speech recognition means including means for converting the speech feature parameter series input from said speech analyzing means into a plurality of speech feature vectors, and means for matching each of the speech feature vectors with the reference vectors of predetermined words in said dictionary means, and outputting a series of a plurality of word candidates used as keywords; and

syntactic analysis means for analyzing the series of the word candidates as the keywords according to syntactic limitations, said syntactic analysis means including means for analyzing a temporal connection relationship of temporally discrete keywords and a syntactic connection relationship thereof, and generating a plurality of sentence candidates.

2. The speech recognition apparatus according to claim 1, wherein said speech analyzing means includes converting means for converting the speech signal to a digital signal, processing the digital signal by means of analysis processing including FFT (Fast Fourier Transform), filter analysis, LPC (Linear Predictive Coding) analysis and cepstrum analysis, and generating the speech feature parameter series.

3. The speech recognition apparatus according to claim 1, wherein said speech recognition means is connected to said dictionary means being storing speech feature patterns corresponding to the keywords, and comprises matching means for matching the speech feature patterns of the dictionary means with each of the speech feature vectors.

4. The speech recognition apparatus according to claim 3, wherein said speech recognition means comprises a speech recognition unit which employs a word spotting method in which the speech feature parameter series generated by said speech analyzing means is matched with the speech feature patterns of said dictionary means at every predetermined frame cycle.

5. The speech recognition apparatus according to claim 4, wherein said speech recognition means matches the speech feature parameter series with the speech feature patterns of said dictionary means at every frame cycle of 8 msec.

6. The speech recognition apparatus according to claim 1, wherein said speech recognition means shifts successively an end point of the input speech feature parameter series by a predetermined time, and matches the speech feature vector obtained in each of start point candidate periods determined with respect to each position of the shifted end point, i.e. in each of periods between the shifted end point and different start points with the speech feature patterns of said dictionary means, in order to perform word recognition and generate a plurality of word candidates.

7. The speech recognition apparatus according to claim 1, wherein said syntactic analysis means includes means for receiving the recognition result from said speech recognition means in the form of temporally discrete word series, performing syntactic analysis by using likelihood of each word, and finding likelihood of each of the sentence candidates by means of evaluation function.

8. The speech recognition apparatus according to claim 7, wherein said syntactic analysis means includes means for outputting preferentially one of the sentence candidates which has a highest likelihood.

9. The speech recognition apparatus according to claim 1, wherein said syntactic analysis means includes means for analyzing a temporal connection relationship of temporally discrete keywords and a syntactic and/or semantic connection relationship thereof, and generating a plurality of semantic expressions corresponding to the analyzed series of the keywords.

10. The speech recognition apparatus according to claim 1, wherein said syntactic analysis means includes grammar/dictionary memory means which stores grammar and dictionary data relating to syntactic/semantic data and temporal relationship of an input sentence, and means for performing syntactic and semantic analysis with use of the grammar/dictionary data of said memory means.

11. The speech recognition apparatus according to claim 1, wherein said syntactic analysis means has sub-sentence candidate buffer means for storing a plurality of sub-sentence candidates produced during analysis.

12. The speech recognition apparatus according to claim 11, wherein said syntactic analysis means includes means for outputting one of the sub-sentence candidates as a sentence candidate.

13. The speech recognition apparatus according to claim 11, wherein said syntactic analysis means comprises means for determining whether one of the word candidates received from said speech recognition means is a top word of a sentence, by referring to the syntactic data stored in said grammar/dictionary memory means, means for generating a semantic expression of one of the sub-sentence candidates as a sub-sentence when said one of the word candidates is the top word of the sentence, and sub-sentence candidate buffer means for storing one of the sub-sentence candidates and its semantic expression.

14. The speech recognition apparatus according to claim 11, further comprising means for predicting a possible word candidate on the basis of syntactic limitations and the sub-sentence candidates under analysis, and using the predicted word candidate for the next input speech recognition processing.

15. The speech recognition apparatus according to claim 14, wherein said predicting means includes means for detecting the word candidates which are connectable to each other from the sub-sentence candidate.

16. The speech recognition apparatus according to claim 1, wherein said speech recognition means is connected to said syntactic analysis means and includes means for outputting word candidates in the form of a word lattice and converting the same to temporally discrete keyword lattices.

17. The speech recognition apparatus according to claim 16, wherein said syntactic analysis means includes means for analyzing a temporary connection relationship of the temporal discrete keywords of the keyword lattices and a syntactic connection relationship thereof, to generate a semantic expression of each of the keyword lattices.

18. A speech recognition apparatus comprising:

speech analyzing means for receiving an external speech signal, analyzing the speech signal and generating a speech feature parameter series;

dictionary memory means for storing a plurality of predetermined words;

speech recognition means including means for converting the speech feature parameter series input from said speech analyzing means into a plurality of speech feature vectors, matching each of the speech feature vectors with the reference vectors of predetermined words in said dictionary memory means, and outputting a series of word candidates;

means for detecting whether or not one of the word candidates is a top word of a sentence and the one of the word candidates being the top word and another following thereto are connectable to each other in time and in grammar, and outputting a combination of the connectable word candidates as a sub-sentence candidate; and

means for determining whether one of the sub-sentence candidates can be established as a sentence, and generating a sentence candidate.

19. A speech recognition apparatus according to claim 18, wherein said means for determining includes means for establishing a sentence when a word candidate corresponding to an end of the sentence is detected.

20. A speech recognition apparatus comprising:

speech recognition means for matching a pattern of an input speech signal with a plurality of prepared dictionary words and outputting a plurality of temporally discrete words as a plurality of keyword candidates; and

syntactic and/or semantic analysis means for checking whether or not the keyword candidates are connectable to each other on the basis of a syntactic and/or semantic rule and outputting a combination of connectable keyword candidates as a sentence candidate.

21. A speech recognition apparatus comprising:

speech analyzing means for receiving an external speech signal, analyzing the speech signal, and generating a speech feature parameter series;

dictionary means being storing a plurality of predetermined words;

speech recognition means including means for converting the speech feature parameter series input from said speech analyzing means into a plurality of speech feature vectors, and means for matching each of the speech feature vectors with the reference vectors of predetermined words in said dictionary means, and outputting a series of a plurality of word candidates used as keywords; and

syntactic analysis means for analyzing the series of the word candidates as the keywords according to syntactic limitations, and generating a plurality of sentence candidates, and wherein

said syntactic analysis means includes grammar/dictionary memory means which stores grammar and dictionary data relating to syntactic/semantic data and temporal relationship of an input sentence, means for determining whether one of the word candidates received from said speech recognition means is a top word of a sentence, by referring to the syntactic data stored in said grammar/dictionary memory means, means for generating a semantic expression of one of the sentence candidates as a sentence when one of the word candidates is the top word of the sentence, and sentence candidate buffer means for storing the sentence and its semantic expression.

22. A speech recognition apparatus comprising:

speech analyzing means for receiving an external speech signal, analyzing the speech signal, and generating a speech feature parameter series;

dictionary means being storing a plurality of predetermined words;

speech recognition means including means for converting the speech feature parameter series input from said speech analyzing means into a plurality of speech feature vectors, and means for matching each of the speech feature vectors with the reference vectors of predetermined words in said dictionary means, and outputting a series of a plurality of word candidates used as keywords;

syntactic analysis means for analyzing the series of the word candidates as the keywords according to syntactic limitations, and generating a sentence candidate, said syntactic analysis means having sub-sentence candidate buffer means for storing a plurality of sub-sentence candidates produced during analysis; and

means for predicting a possible word candidate on the basis of syntactic limitations and the sub-sentence candidates under analysis, and using the predicted word candidate for the next input speech recognition processing.