Method and apparatus for a speech recognition system language model that integrates a finite state grammar probability and an N-gram probability

• US 6,154,722 A
• Filed: 12/18/1997
• Issued: 11/28/2000
• Est. Priority Date: 12/18/1997
• Status: Expired due to Term

• Current Assignee: Apple Inc.
• Original Assignee: Apple Computer Incorporated (Apple Inc.)
• Inventors: Bellegarda, Jerome R.
• Primary Examiner(s): Zele, Krista
• Assistant Examiner(s): Opsasnick, Michael N.
• Application Number:
  US08/993,939
• Time in Patent Office: 1,076 Days
• Field of Search: 704/257, 704/256, 704/255
• US Class Current: 704/257
• CPC Class Codes: G10L 15/193 Formal grammars, e.g. finite state automata, context free grammars or word networks
• CPC Class Codes: G10L 15/197 Probabilistic grammars, e.g. word n-grams

Abstract

A method and an apparatus for a speech recognition system that uses a language model based on an integrated finite state grammar probability and an n-gram probability are provided. According to one aspect of the invention, speech signals are received into a processor of a speech recognition system. The speech signals are processed using a speech recognition system hosting a language model. The language model is produced by integrating a finite state grammar probability and an n-gram probability. In the integration, the n-gram probability is modified based on information provided by the finite state grammar probability; thus, the finite state grammar probability is subordinate to the n-gram probability. The language model is used by a decoder along with at least one acoustic model to perform a hypothesis search on an acoustic sequence to provide a word sequence output. The word sequence generated is representative of the received speech signals.

Claims

1. A method for recognizing speech comprising:

receiving speech signals into a processor;

processing the received speech signals using a language model produced by integrating a finite state grammar probability and an n-gram probability with the finite state grammar probability being subordinated to the n-gram probability, wherein the language model comprises a probability that is specified by the equation ##EQU4## where, w_q is the word about to be predicted, H_q is the admissible history for this particular word, G is the probabilistic finite state grammar, S_q is the present sentence up to w_q, and V is the vocabulary to which w_q belongs; and

generating a word sequence representative of the received speech signals.

2. The method of claim 1, wherein the language model is used with at least one acoustic model to perform a hypothesis search on an acoustic sequence to provide a word sequence output.

3. The method of claim 2, wherein the at least one acoustic model is based on a hidden Markov model paradigm, wherein the at least one acoustic model comprises one model for each of at least one phonemes.

4. An apparatus for speech recognition comprising:

an input for receiving speech signals into a processor;

a processor configured to recognize the received speech signals using a language model produced by integrating a finite state grammar paradigm and an n-gram paradigm, with the finite state grammar paradigm being subordinated to the n-gram paradigm, wherein the language model comprises a probability that is specified by the equation ##EQU5## where, w_q is the word about to be predicted, H_q is the admissible history for this particular word, G is the probabilistic finite state grammar, S_q is the present sentence up to w_q, and V is the vocabulary to which w_q belongs; and

an output for providing a word sequence representative of the received speech signals.

5. The apparatus of claim 4, wherein the processor is further configured to control a hypothesis search by a decoder, the decoder using inputs from the language model and from at least one acoustic model to execute the hypothesis search.

6. The apparatus of claim 5, wherein the at least one acoustic model is based on a hidden Markov model paradigm.

7. A speech recognition process comprising a statistical learning technique that uses a language model produced by integrating a finite state grammar probability and an n-gram probability with the finite state grammar probability being subordinated to the n-gram probability, wherein the language model comprises a probability that is specified by the equation ##EQU6## where, w_q is the word about to be predicted, H_q is the admissible history for this particular word, G is the probabilistic finite state grammar, S_q is the present sentence up to w_q, and V is the vocabulary to which w_q belongs.

8. The process of claim 7, wherein at least one acoustic model is used, the at least one acoustic model used with the language model in executing a hypothesis search on an acoustic sequence to provide a word sequence output.

9. A computer readable medium containing executable instructions which, when executed in a processing system, causes the system to perform a method for recognizing speech comprising:

receiving a speech signal into a processor;

processing the received speech signal using a language model produced by integrating a finite state grammar probability and an n-gram probability with the finite state gammar probability being subordinated to the n-gram probability, wherein the language model comprises a probability that is specified by the equation ##EQU7## where, w_q is the word about to be predicted, H_q is the admissible history for this particular word, G is the probabilistic finite state grammar, S_q is the present sentence up to w_q, and V is the vocabulary to which w_q belongs; and

generating a word sequence representative of the received speech signal.

10. The computer readable medium of claim 9, wherein the language model is used with at least one acoustic model to perform a hypothesis search on an acoustic sequence to provide a word sequence output.

11. A method for generating a language model for use in a speech recognition system comprising:

integrating a finite state grammar probability and an n-gram probability with the finite state grammar probability being subordinated to the n-gram probability, wherein the language model comprises a probability that is specified by the equation ##EQU8## where, w_q is the word about to be predicted, H_q is the admissible history for this particular word, G is the probabilistic finite state grammar, S_q is the present sentence up to w_q, and V is the vocabulary to which w_q belongs.

12. The method of claim 11, further comprising using the language model along with at least one acoustic model to perform a hypothesis search on an acoustic sequence to provide a word sequence output.

13. The method of claim 12, wherein the at least one acoustic model is based on a hidden Markov model paradigm, wherein the at least one acoustic model comprises one model for each of at least one phonemes.