System and method for cantonese speech recognition using an optimized phone set

• US 20040193418A1
• Filed: 03/24/2003
• Published: 09/30/2004
• Est. Priority Date: 03/24/2003
• Status: Active Grant

• Current Assignee: Ironworks Patents LLC, Sony Corporation (Sony Group Corp.)
• Original Assignee: SONY CORPORATION AND SONY ELECTRONICS INC.
• Inventors: Menendez-Pidal, Xavier, Emonts, Michael, Olorenshaw, Lex
• Granted Patent: US 7,353,172 B2
• Time in Patent Office: Days
• US Class Current: 704/256
• CPC Class Codes: G10L 15/187 Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units

Abstract

The present invention comprises a system and method for implementing a Cantonese speech recognizer with an optimized phone set, and may include a recognizer configured to compare input speech data to phone strings from a vocabulary dictionary that is implemented according to an optimized Cantonese phone set. The optimized Cantonese phone set may be implemented with a phonetic technique to separately include consonantal phones and vocalic phones. For reasons of system efficiency, the optimized Cantonese phone set may preferably be implemented in a compact manner to include only a minimum required number of consonantal phones and vocalic phones to accurately represent Cantonese speech during the speech recognition procedure.

Claims

1. A system for performing a speech recognition procedure, comprising:

a recognizer configured to compare input speech data to phone strings from a vocabulary dictionary that is implemented according to an optimized phone set, said optimized phone set being implemented with a phonetic technique to separately provide consonantal phones and vocalic phones, said optimized phone set being implemented in a compact manner to include only a minimum required number of said consonantal phones and said vocalic phones; and

a processor configured to control said recognizer to thereby perform said speech recognition procedure.

2. The system of claim 1 wherein said input speech data includes Cantonese language data, said optimized phone set being compactly configured to accurately represent said Cantonese language data.

3. The system of claim 1 wherein said recognizer and said processor are implemented as part of a consumer electronics device.

4. The system of claim 1 wherein said optimized phone set conserves processing resources and memory resources while performing said speech recognition procedure.

5. The system of claim 1 wherein said optimized phone set reduces training requirements for performing a recognizer training procedure to initially implement said recognizer.

6. The system of claim 1 wherein said phone strings each include a different series of phones from said optimized phone set, each of said phone strings corresponding to a different word from said vocabulary dictionary.

7. The system of claim 6 wherein said recognizer compares said input speech data to Hidden Markov Models for said phone strings from said vocabulary dictionary to thereby select a recognized word during said speech recognition procedure.

8. The system of claim 1 wherein said optimized phone set includes phones b, d, g, p, t, k, m, n, ng, f, l, h, z, c, s, w, j, cl, sil, aa, i, u, e, o, yu, oe, eo, a, eu, aai, aau, ai, au, ei, oi, ou, eoi, ui, and iu.

9. The system of claim 1 wherein said optimized phone set includes consonantal phones b, d, g, p, t, k, m, n, ng, f, l, h, z, c, s, w, and j.

10. The system of claim 1 wherein said optimized phone set includes a closure phone “

cl” and

a silence phone “

sil”

.

11. The system of claim 1 wherein said optimized phone set includes vocalic phones aa, i, u, e, o, yu, oe, eo, a, eu, aai, aau, ai, au, ei, oi, ou, eoi, ui, and iu.

12. The system of claim 1 wherein said optimized phone set represents certain diphthongs by utilizing unified diphthong phones to thereby conserve processing resources and memory resources while providing greater accuracy characteristics for said speech recognition procedure.

13. The system of claim 12 wherein said optimized phone set includes unified diphthong phones eu, aai, aau, ai, au, ei, oi, ou, eoi, ui, and iu.

14. The system of claim 1 wherein said optimized phone set represents a certain lip rounding by utilizing a separate lip rounding phone “

w”

after a consonantal phone “

g”

.

15. The system of claim 1 wherein said optimized phone set represents a certain lip rounding by utilizing a separate lip rounding phone “

w”

after a consonantal phone “

k”

.

16. The system of claim 1 wherein said input speech data includes a syllable-initial context in which a stop is located at a beginning of a syllable, said optimized phone set responsively utilizing an appropriate consonant phone “

p”

, “

t”

, or “

k”

in said syllable-initial context to represent a corresponding consonant and a preceding closure.

17. The system of claim 1 wherein said input speech data includes a syllable-final/midphrase context in which a stop is located at an end of a word in a middle of a phrase, said optimized phone set responsively utilizing an appropriate consonant phone “

p”

, “

t”

, or “

k”

in said syllable-final/midphrase context to represent a corresponding consonant and a preceding closure.

18. The system of claim 1 wherein said input speech data includes a syllable-final/phrase-end context in which a stop is located at an end of a word at an end of a phrase, said optimized phone set responsively utilizing a same identical closure phone “

cl”

in said syllable-final/phrase-end context to represent either “

p”

, “

t”

, or “

k”

consonants as a closure only without any subsequent releasing consonant sound.

19. The system of claim 1 wherein said input speech data includes a syllable-initial context in which a first stop is located at a beginning of a syllable, a syllable-final/midphrase context in which a second stop is located at an end of a first word in a middle of a phrase, and a syllable-final/phrase-end context in which a third stop is located at an end of a second word at an end of said phrase, said optimized phone set utilizing an appropriate consonant phone “

b”

, “

d”

, “

g”

, “

p”

, “

t”

, or “

k”

in said syllable-initial context to represent a corresponding consonant and a preceding closure, said optimized phone set responsively utilizing said appropriate consonant phone “

p”

, “

t”

, or “

k”

in said syllable-final/midphrase context to represent said corresponding consonant and said preceding closure, said optimized phone set responsively utilizing a same identical closure phone “

cl”

in said syllable-final/phrase-end context to represent either “

p”

, “

t”

, or “

k”

as a closure only without any subsequent releasing consonant sound.

20. The system of claim 1 wherein said consonantal phones and said vocalic phones from said optimized phone set are combined to represent syllables from a Cantonese language system.

21. A method for performing a speech recognition procedure, comprising the steps of:

configuring a recognizer to compare input speech data to phone strings from a vocabulary dictionary that is implemented according to an optimized phone set, said optimized phone set being implemented with a phonetic technique to separately provide consonantal phones and vocalic phones, said optimized phone set being implemented in a compact manner to include only a minimum required number of said consonantal phones and said vocalic phones; and

controlling said recognizer with a processor to thereby perform said speech recognition procedure.

22. The method of claim 21 wherein said input speech data includes Cantonese language data, said optimized phone set being compactly configured to accurately represent said Cantonese language data.

23. The method of claim 21 wherein said recognizer and said processor are implemented as part of a consumer electronics device.

24. The method of claim 21 wherein said optimized phone set conserves processing resources and memory resources while performing said speech recognition procedure.

25. The method of claim 21 wherein said optimized phone set reduces training requirements for performing a recognizer training procedure to initially implement said recognizer.

26. The method of claim 21 wherein said phone strings each include a different series of phones from said optimized phone set, each of said phone strings corresponding to a different word from said vocabulary dictionary.

27. The method of claim 26 wherein said recognizer compares said input speech data to Hidden Markov Models for said phone strings from said vocabulary dictionary to thereby select a recognized word during said speech recognition procedure.

28. The method of claim 21 wherein said optimized phone set includes phones b, d, g, p, t, k, m, n, ng, f, l, h, z, c, s, w, j, cl, sil, aa, i, u, e, o, yu, oe, eo, a, eu, aai, aau, ai, au, ei, oi, ou, eoi, ui, and iu.

29. The method of claim 21 wherein said optimized phone set includes consonantal phones b, d, g, p, t, k, m, n, ng, f, l, h, z, c, s, w, and j.

30. The method of claim 21 wherein said optimized phone set includes a closure phone “

cl” and

a silence phone “

sil”

.

31. The method of claim 21 wherein said optimized phone set includes vocalic phones aa, i, u, e, o, yu, oe, eo, a, eu, aai, aau, ai, au, ei, oi, ou, eoi, ui, and iu.

32. The method of claim 21 wherein said optimized phone set represents certain diphthongs by utilizing unified diphthong phones to thereby conserve processing resources and memory resources while providing greater accuracy characteristics for said speech recognition procedure.

33. The method of claim 32 wherein said optimized phone set includes unified diphthong phones eu, aai, aau, ai, au, ei, oi, ou, eoi, ui, and iu.

34. The method of claim 21 wherein said optimized phone set represents a certain lip rounding by utilizing a separate lip rounding phone “

w”

after a consonantal phone “

g”

.

35. The method of claim 21 wherein said optimized phone set represents a certain lip rounding by utilizing a separate lip rounding phone “

w”

after a consonantal phone “

k”

.

36. The method of claim 21 wherein said input speech data includes a syllable-initial context in which a stop is located at a beginning of a syllable, said optimized phone set responsively utilizing an appropriate consonant phone “

b”

, “

d”

, “

g”

, “

p”

, “

t”

, or “

k”

in said syllable-initial context to represent a corresponding consonant and a preceding closure.

37. The method of claim 21 wherein said input speech data includes a syllable-final/midphrase context in which a stop is located at an end of a word in a middle of a phrase, said optimized phone set responsively utilizing an appropriate consonant phone “

p”

, “

t”

, or “

k”

in said syllable-final/midphrase context to represent a corresponding consonant and a preceding closure.

38. The method of claim 21 wherein said input speech data includes a syllable-final/phrase-end context in which a stop is located at an end of a word at an end of a phrase, said optimized phone set responsively utilizing a same identical closure phone “

cl”

in said syllable-final/phrase-end context to represent either “

p”

, “

t”

, or “

k”

as a closure only without any subsequent releasing consonant sound.

39. The method of claim 21 wherein said input speech data includes a syllable-initial context in which a first stop is located at a beginning of a syllable, a syllable-final/midphrase context in which a second stop is located at an end of a first word in a middle of a phrase, and a syllable-final/phrase-end context in which a third stop is located at an end of a second word at an end of said phrase, said optimized phone set utilizing an appropriate consonant phone “

b”

, “

d”

, “

g”

, “

p”

, “

t”

, or “

k”

in said syllable-initial context to represent a corresponding consonant and a preceding closure, said optimized phone set responsively utilizing an appropriate consonant phone “

p”

, “

t”

, or “

k”

in said syllable-final/midphrase context to represent said corresponding consonant and a preceding closure, said optimized phone set responsively utilizing a same identical closure phone “

cl”

in said syllable-final/phrase-end context to represent either “

p”

, “

t”

, or “

k”

as a closure only without any subsequent releasing consonant sound.

40. The method of claim 21 wherein said consonantal phones and said vocalic phones from said optimized phone set are combined to represent syllables from a Cantonese language system.

41. The method of claim 21 wherein said optimized phone set represents various sounds of a Cantonese language without utilizing corresponding tonal information as part of different phones in said optimized phone set.

42. A computer-readable medium comprising program instructions for performing a speech recognition, by performing the steps of:

configuring a recognizer to compare input speech data to phone strings from a vocabulary dictionary that is implemented according to an optimized phone set, said optimized phone set being implemented with a phonetic technique to separately provide consonantal phones and vocalic phones, said optimized phone set being implemented in a compact manner to include only a minimum required number of said consonantal phones and said vocalic phones; and

controlling said recognizer with a processor to thereby perform said speech recognition procedure.

43. A system for performing a speech recognition procedure, comprising the steps of:

means for comparing input speech data to phone strings from a vocabulary dictionary that is implemented according to an optimized phone set, said optimized phone set being implemented with a phonetic technique to separately provide consonantal phones and vocalic phones, said optimized phone set being implemented in a compact manner to include only a minimum required number of said consonantal phones and said vocalic phones; and

means for controlling said means for comparing to thereby perform said speech recognition procedure.