DISCRIMINATIVE TRAINING OF AUTOMATIC SPEECH RECOGNITION MODELS WITH NATURAL LANGUAGE PROCESSING DICTIONARY FOR SPOKEN LANGUAGE PROCESSING

• US 20170169813A1
• Filed: 12/14/2015
• Published: 06/15/2017
• Est. Priority Date: 12/14/2015
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Suzuki, Masayuki, Itoh, Nobuyasu, Kurata, Gakuto, Nagano, Tohru
• Granted Patent: US 10,140,976 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 40/242 Dictionaries
• CPC Class Codes: G06F 40/284 Lexical analysis, e.g. tokenisation or collocates
• CPC Class Codes: G10L 15/01 Assessment or evaluation of speech recognition systems
• CPC Class Codes: G10L 15/063 Training
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/197 Probabilistic grammars, e.g. word n-grams
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 2015/0635 updating or merging of old and new templates; Mean values; Weighting
• CPC Class Codes: G10L 2015/0638 Interactive procedures

Abstract

Methods and systems for language processing includes training one or more automatic speech recognition models using an automatic speech recognition dictionary. A set of N automatic speech recognition hypotheses for an input is determined, based on the one or more automatic speech recognition models, using a processor. A best hypothesis is selected using a discriminative language model and a list of relevant words. Natural language processing is performed on the best hypothesis.

Claims

1. A method for language processing, comprising:

training one or more automatic speech recognition models using an automatic speech recognition dictionary;

determining a set of N automatic speech recognition hypotheses that characterize a spoken input, based on the one or more automatic speech recognition models, using a processor;

selecting a hypothesis from the set of N automatic speech recognition hypotheses using a discriminative language model and a list of relevant words according to an error rate of relevant words in each hypothesis; and

performing natural language processing on the best hypothesis.

2. The method for language processing of claim 1, further comprising determining the list of relevant words using natural language processing training data and the automatic speech recognition dictionary by trimming out words from the automatic speech recognition dictionary that are unlikely to be relevant to natural language processing.

3. The method for language processing of claim 2, wherein determining the list of relevant words comprises concatenating all words in the natural language processing training data to generate raw natural language processing text.

4. The method for language processing of claim 3, further comprising tokenizing the raw natural language processing text using the automatic speech recognition dictionary.

5. The method for language processing of claim 4, further comprising collecting tokenized words that appear more than a threshold number of times as relevant words.

6. The method for language processing of claim 5, further comprising adding entries of a first natural language processing dictionary to the collected tokenized words to form the relevant word list.

7. The method for language processing of claim 1, wherein selecting the best hypothesis comprises determining a relevant word error rate for each hypothesis that considers only errors in words of the respective hypothesis that are on the relevant word list using the discriminative language model.

8. The method for language processing of claim 7, further comprising selecting the hypothesis having the lowest relevant word error rate.

9. The method for language processing of claim 1, wherein training the one or more automatic speech recognition models comprises training an acoustic model and a language mode.

10. A computer readable storage medium comprising a computer readable program for language processing, wherein the computer readable program when executed on a computer causes the computer to perform the steps of claim 1.

11. A method for language processing, comprising:

training one or more automatic speech recognition models using an automatic speech recognition dictionary;

determining a set of N automatic speech recognition hypotheses that characterize a spoken input, based on the one or more automatic speech recognition models, using a processor, comprising;

concatenating all words in the natural language processing training data to generate raw natural language processing text;

tokenizing the raw natural language processing text using the automatic speech recognition dictionary;

collecting tokenized words that appear more than a threshold number of times; and

adding entries of a natural language processing dictionary to the collected tokenized words to form the relevant word list;

selecting a hypothesis from the set of N automatic speech recognition hypotheses using a discriminative language model and a list of relevant words according to an error rate of relevant words in each hypothesis, comprising;

determining a relevant word error rate for each hypothesis that considers only words that are on the relevant word list using the discriminative language model; and

selecting the hypothesis having the lowest relevant word error rate; and

performing natural language processing on the best hypothesis.

12. A system for language processing, comprising:

an automatic speech recognition module comprising a processor configured to train one or more automatic speech recognition models using an automatic speech recognition dictionary, to determine a set of N automatic speech recognition hypotheses that characterize a spoken input based on the one or more automatic speech recognition models, and to select a hypothesis from the set of N automatic speech recognition hypotheses using a discriminative language model and a list of relevant words according to an error rate of relevant words in each hypothesis; and

a natural language processing module configured to perform natural language processing on the best hypothesis.

13. The system of claim 12, wherein the automatic speech recognition module is further configured to determine the list of relevant words using natural language processing training data and the automatic speech recognition dictionary by trimming out words from the automatic speech recognition dictionary that are unlikely to be relevant to natural language processing.

14. The system for language processing of claim 13, wherein the automatic speech recognition module is further configured to concatenate all words in the natural language processing training data to generate raw natural language processing text.

15. The system for language processing of claim 14, wherein the automatic speech recognition module is further configured to tokenize the raw natural language processing text using the automatic speech recognition dictionary.

16. The system for language processing of claim 15, wherein the automatic speech recognition module is further configured to collect tokenized words that appear more than a threshold number of times as relevant words.

17. The system for language processing of claim 16, wherein the automatic speech recognition module is further configured to add entries of a first natural language processing dictionary to the collected tokenized words to form the relevant word list.

18. The system for language processing of claim 12, wherein the automatic speech recognition module is further configured to determine a relevant word error rate for each hypothesis that considers only errors in words of the respective hypothesis that are on the relevant word list using the discriminative language model.

19. The system for language processing of claim 18, wherein the automatic speech recognition module is further configured to select the hypothesis having the lowest relevant word error rate.

20. The system for language processing of claim 12, wherein the automatic speech recognition module is further configured to train an acoustic model and a language mode.