Voice recognition method with automatic correction

• US 20060015338A1
• Filed: 09/19/2003
• Published: 01/19/2006
• Est. Priority Date: 09/24/2002
• Status: Abandoned Application

• Current Assignee: Thales SA
• Original Assignee: Thales SA
• Inventors: Poussin, Gilles
• Application Number:
  US10/527,132
• Publication Number:
  US 20060015338A1
• Time in Patent Office: Days
• US Class Current: 704/251
• CPC Class Codes: G10L 15/193 Formal grammars, e.g. finite state automata, context free grammars or word networks
• CPC Class Codes: G10L 15/22 Procedures used during a speech recognition process, e.g. man-machine dialogue

Abstract

The present invention relates to a method of voice recognition with automatic correction in voice recognition systems with constrained syntax. It comprises in particular a step of processing said speech signal delivering a signal in a compressed form, a step of recognizing patterns so as to search, on the basis of a syntax formed of a set of phrases which represent the set of possible paths between a set of words prerecorded during a prior phase, for a phrase of said syntax that is the closest to said signal in its compressed form, the storage of the signal in its compressed form, the generation of a new syntax in which the path corresponding to said phrase determined during the earlier recognition step is precluded, the repetition of the step of recognizing patterns so as to search, on the basis of the new syntax, for another phrase which is the closest to said stored signal.

Claims

1. A method of voice recognition of a speech signal uttered by a speaker with automatic correction, steps of:

processing said speech signal and delivering a signal in a compressed form;

recognizing patterns so as to search, on the basis of a syntax formed of a set of phrases which represent the set of possible paths between a set of words prerecorded during a prior phase, for a phrase of said syntax that is the closest to said signal in its compressed form;

storing the signal in its compressed form,

generating a new syntax in which the path corresponding to said phrase determined during the earlier recognition step is precluded,

repeating the step of recognizing patterns so as to search, on the basis of the new syntax, for another phrase that is the closest to said stored signal.

2. The method of voice recognition as claimed in claim 1, in which the new syntax is obtained by reorganizing the earlier syntax in such a way as to particularize said path corresponding to the phrase determined during the earlier recognition step, then eliminating this path.

3. The method of voice recognition as claimed in claim 2, in which said reorganization is effected by traversing the earlier syntax as a function of the words of said phrase and formation in the course of this traversal of the path specific to said phrase.

4. The method of voice recognition as claimed in claim 1, characterized in that wherein the search for a new phrase is repeated systematically to anticipate the correction.

5. The method of voice recognition as claimed in claim 4, wherein each new phrase recognized is proposed to the speaker on the request thereof.

6. The method of voice recognition as claimed in claim 4, wherein the search for a new phrase is halted by validation of a phrase recognized by the speaker.

7. The method of voice recognition as claimed in claim 1, characterized in that wherein the processing step comprises:

digitizing and chopping into a string of time frames of said acoustic signal,

a phase of parameterization of time frames containing the speech so as to obtain, per frame, a vector of parameters in the frequency domain, the whole set of these parameter vectors forming said signal in its compressed form.

8. The method of voice recognition as claimed in claim 7, wherein the pattern recognition calls upon an algorithm of DTW type.

9. The method of voice recognition as claimed in claim 7, wherein the pattern recognition calls upon an algorithm of HMM type.

10. The method of voice recognition as claimed in claim 2, wherein the search for a new phrase is repeated systematically to anticipate the correction.

11. The method of voice recognition as claimed in claim 3, wherein the search for a new phrase is repeated systematically to anticipate the correction.

12. The method of voice recognition as claimed in claim 5, wherein the search for a new phrase is halted by validation of a phrase recognized by the speaker.

13. The method of voice recognition as claimed in claim 2, wherein the processing step comprises:

digitizing and chopping into a string of time frames of said acoustic signal,

a phase of parameterization of time frames containing the speech so as to obtain, per frame, a vector of parameters in the frequency domain, the whole set of these parameter vectors forming said signal in its compressed form.

14. The method of voice recognition as claimed in claim 3, wherein the processing step comprises:

digitizing and chopping into a string of time frames of said acoustic signal,

a phase of parameterization of time frames containing the speech so as to obtain, per frame, a vector of parameters in the frequency domain, the whole set of these parameter vectors forming said signal in its compressed form.

15. The method of voice recognition as claimed in claim 4, wherein the processing step comprises:

digitizing and chopping into a string of time frames of said acoustic signal,

a phase of parameterization of time frames containing the speech so as to obtain, per frame, a vector of parameters in the frequency domain, the whole set of these parameter vectors forming said signal in its compressed form.

16. The method of voice recognition as claimed in claim 5, wherein the processing step comprises:

digitizing and chopping into a string of time frames of said acoustic signal,

a phase of parameterization of time frames containing the speech so as to obtain, per frame, a vector of parameters in the frequency domain, the whole set of these parameter vectors forming said signal in its compressed form.