Automatic Speech Recognition System and Method

• US 20080114595A1
• Filed: 12/28/2004
• Published: 05/15/2008
• Est. Priority Date: 12/28/2004
• Status: Active Grant

• Current Assignee: Nuance Communications, Inc. (Microsoft Corporation)
• Original Assignee: Loquendo SpA (Microsoft Corporation)
• Inventors: Colibro, Daniele, Vair, Claudio
• Granted Patent: US 7,912,713 B2
• Time in Patent Office: Days
• US Class Current: 704/236
• CPC Class Codes: G10L 15/08 Speech classification or search
• CPC Class Codes: G10L 15/144 Training of HMMs

Abstract

An automatic speech recognition method for identifying words from an input speech signal includes providing at least one hypothesis recognition based on the input speech signal, the hypothesis recognition being an individual hypothesis word or a sequence of individual hypothesis words, and computing a confidence measure for the hypothesis recognition, based on the input speech signal, wherein computing a confidence measure includes computing differential contributions to the confidence measure, each as a difference between a constrained acoustic score and an unconstrained acoustic score, weighting each differential contribution by applying thereto a cumulative distribution function of the differential contribution, so as to make the distributions of the confidence measures homogeneous in terms of rejection capability, as the language, vocabulary and grammar vary, and computing the confidence measure by averaging the weighted differential contributions.

Claims

1-17. -17. (canceled)

18. An automatic speech recognition method for identifying words from an input speech signal, comprising:

providing at least one hypothesis recognition based on said input speech signal, said hypothesis recognition being an individual hypothesis word or a sequence of individual hypothesis words; and

computing a confidence measure for said hypothesis recognition, based on said input speech signal, said computing a confidence measure comprising;

computing differential contributions (C_i,t) to said confidence measure, each as a difference between a constrained acoustic score (SCV_i,t) and an unconstrained acoustic score (SCL_t); and

computing said confidence measure based on said differential contributions (C_i,t), comprising;

computing said confidence measure based on said differential contributions (C_i,t) comprising;

weighting each differential contribution (C_i,t) with a cumulative distribution function (F<sub>ξ

i</sub>) of the differential contribution (C_i,t).

19. The automatic speech recognition method as claimed in claim 18, wherein said cumulative distribution functions (F_ξ

_i) of the differential contributions (C_i,t) are such that distributions of the confidence measures are homogeneous in terms of rejection capability, as at least one of language, vocabulary and grammar varies.

20. An automatic speech recognition method as claimed in claim 18, wherein each differential contribution (C_i,t) is computed according to the following equation:

C_i,t=F<sub>ξ

i</sub>(SCV_i,t−

SCL_t)

wherein C_i,t is said differential contribution, (F<sub>ξ

i</sub>) is said cumulative distribution function, SCV_i,t is said constrained acoustic score for acoustic state i of a hidden Markov model which models a given acoustic-phonetic unit, which, in turn, models a corresponding speech phenomena at input speech frame t, and SCL_t is said unconstrained acoustic score at input speech frame t.

21. The automatic speech recognition method as claimed in claim 20, wherein said constrained acoustic score (SCV_i,t) is computed according to the following equation:

SCV_i,t=log P(S_i|o_t)

wherein P(S_i*o_t) is a likelihood of emission of acoustic states S_i, given an observation vector o_t extracted from the input speech signal at input speech frame t.

22. The automatic speech recognition method as claimed in claim 21, wherein said unconstrained acoustic score (SCL_t) is computed according to the following equation:

23. The automatic speech recognition method as claimed in claim 22, wherein each differential contribution (C_i,t) is computed according to the following equation:

24. The automatic speech recognition method as claimed in claim 18, wherein said confidence measure is computed as a temporal average of said weighted differential contributions (C_i,t) over the sequence of recognized acoustic states (S_i*) of elementary acoustic-phonetic units forming said recognition.

25. The automatic speech recognition method as claimed in claim 23, wherein said confidence measure is computed according to the following equation:

26. The automatic speech recognition method as claimed in claim 21, wherein said emission likelihood is a posteriori likelihood computed from a likelihood based on a hidden Markov model.

27. The automatic speech recognition method as claimed in claim 21, wherein said emission likelihood is an a posteriori likelihood computed by an artificial neural network of a hybrid hidden Markov model-neural network system.

28. The automatic speech recognition method as claimed in claim 21, wherein said emission likelihood is computed by a hybrid hidden Markov model-neural network system according to the following equation:

29. The automatic speech recognition method as claimed in claim 28, wherein said differential contribution (C_i,t) is computed as follows:

30. An automatic speech recognition method as claimed in claim 29, wherein said confidence measure is computed as follows:

31. The automatic speech recognition method as claimed in claim 18, wherein each cumulative distribution function (F_ξ

_i) is based on a probability density function with one of a Gaussian shape, a binomial shape, a Poisson shape, and a Gamma shape.

32. The automatic speech recognition method as claimed in claim 18, wherein each cumulative distribution function (F_ξ

_i) is trained from training speech data using fitting methods.

33. An automatic speech recognition system for identifying words from an input speech signal, comprising:

a search engine configured to provide at least one hypothesis recognition based on said input speech signal, said hypothesis recognition being an individual hypothesis word or a sequence of individual hypothesis words; and

a confidence measure module configured to compute a confidence measure for said hypothesis recognition based on said input speech signal;

said confidence measure module being configured to compute said confidence measure as claimed in claim 18.

34. A computer program product comprising a computer program code capable, when loaded in a processing system, of implementing the method according to claim 18.