Method and device for speech recognition

• US 5,694,520 A
• Filed: 01/11/1996
• Issued: 12/02/1997
• Est. Priority Date: 06/29/1994
• Status: Expired due to Term

• Current Assignee: Intellectual Ventures I LLC (Intellectual Ventures LLC)
• Original Assignee: Telia AB (Government of Norway)
• Inventors: Lyberg, Bertil
• Primary Examiner(s): Tung, Kee M.
• Application Number:
  US08/532,823
• Time in Patent Office: 691 Days
• Field of Search: 395/2.86, 395/2.6, 395/2.63, 395/2.64, 395/2.66, 395/2.09, 395/2.14, 395/2.16, 395/2.2
• US Class Current: 704/254
• CPC Class Codes: G10L 15/07 to the speaker
• CPC Class Codes: G10L 15/1807 using prosody or stress
• CPC Class Codes: G10L 15/187 Phonemic context, e.g. pronunciation rules, phonotactical constraints or phoneme n-grams
• CPC Class Codes: G10L 15/19 Grammatical context, e.g. disambiguation of the recognition hypotheses based on word sequence rules
• CPC Class Codes: G10L 2015/025 Phonemes, fenemes or fenones being the recognition units
• CPC Class Codes: G10L 25/90 Pitch determination of speech signals

Abstract

A method and device for recognizing dialectal variations in a language. From an incoming speech is on one hand a speech recognition procedure being performed, and on the other hand the fundamental tone curve being extracted. Out of the speech recognition is created an allophone string which together with the fundamental tone curve is used for the detecting of the maximun and minimum values of the fundamental tone. The recognized speech is compared with a lexicon with orthography and transcription for the finding of suitable word candidates. The found word candidates are further analyzed regarding syntax. This in mentioned way found syntactical and lexical information is used for creating a model of the speech. The fundamental tone outline of the model and the fundamental tone of the speech are compared, at which the maximun and minimum values of the fundamental tones are appointed and a difference between the model and the speech are obtained. The difference is after that influencing the model which is brought to correspond to the given speech. The in mentioned way modelled model is then used for the speech recognition, at which an increased possibility to understand the different dialects of a language in an artificial way is achieved.

Claims

1. A method for recognizing spoken language comprising the steps of:

identifying a number of phonemes from a segment of input speech;

interpreting the phonemes as possible word combinations to establish a model of the speech with word and sentence accents according to a standardized pattern;

determining the fundamental tone curve of the input speech;

determining the maximum and minimum values of the fundamental tone curve of the input speech and their respective positions;

determining the maximum and minimum values of the fundamental tone curve of the speech model;

comparing the fundamental tone curve of the input speech and the fundamental tone curve of the speech model to identify a time difference between the maximum and minimum values of the fundamental tone curve of the incoming speech in relation to the maximum and minimum values of the fundamental tone curve of the speech model;

adjusting the intonation pattern of the speech model utilizing the identified time difference to modify the speech model to conform with the dialectal characteristics of the input speech.

2. Method according to claim 1, characterized by the time difference being appointed in relation to a reference, preferably a CV-limit, where C indicate a consonant and V a vowel.

3. Method according to claim 2 characterized by the fundamental tone outline being based on lexical and syntactical information, and information about orthography and fonetic transcription.

4. Method according to claim 3, characterized by the speech on the basis of FO-events being classified in dialect categories depending on stored information descriptions.

5. Method according to claim 2, characterized by the transcription containing a lexical abstract accent information type stressed syllable, tonal word accents, accent I and accent II as well as location of secondary accent, i.e., information normally provided by dictionary.

6. Method according to claim 2, characterized by the speech on the basis of FO-events being classified in dialect categories depending on stored information descriptions.

7. Method according to claim 1, characterized by the fundamental tone outline being based on lexical and syntactical information, and information about orthography and fonetic transcription.

8. Method according to claim 7, characterized by the transcription containing a lexical abstract accent information type stressed syllable, tonal word accents, accent I and accent II as well as location of secondary accent, i.e., information normally provided by dictionary.

9. Method according to claim 7, characterized by the speech on the basis of FO-events being classified in dialect categories depending on stored information descriptions.

10. Method according to claim 1, characterized by the transcription containing a lexical abstract accent information type stressed syllable, tonal word accents accent I and accent II as well as location of secondary accent, i.e. information normally provided by dictionary.

11. Method according to claim 10, characterized by the speech on the basis of FO-events being classified in dialect categories depending on stored information descriptions.

12. Method according to claim 1, characterized by the speech on the basis of FO-events being classified in dialect categories depending on stored information descriptions.

13. A device for recognizing spoken language comprising:

speech recognition means for identifying a number of phonemes from a segment of input speech;

interpretation means for interpreting the phonemes as possible word combinations to establish a model of speech having word and sentence accents according to a standardized pattern;

extraction means for extracting a fundamental tone curve of the input speech;

first analyzing means for determining the maximum and minimum values of the fundamental tone curve and their respective positions;

second analyzing means for determining the maximum and minimum values of the fundamental tone curve of the speech model and their respective positions;

comparison means for comparing the input speech with the speech model to identify a time difference between the occurrence of the maximum and minimum values of the fundamental tone curve of the incoming speech in relation to the maximum and minimum values of the fundamental tone curve of the speech model;

correction means for adjusting the intonation pattern of the speech model utilizing the identified time difference, to modify the speech model to conform with the dialectal characteristics of the input speech.

14. Device according to claim 13, characterized by the model instrument being arranged to, from a lexicon with orthography and transcription, create a model of the speech.

15. Device according to claim 14, characterized by the model instrument being arranged to analyze the lexical information regarding syntax at which the model is brought to correspond to one in the language standardized pattern.

16. Device according to claim 15, characterized by the device being arranged to categorize the speech in different dialect categories on the basis of stored intonation descriptions in the device.

17. Device according to claim 14, characterized by the device being arranged to categorize the speech in different dialect categories on the basis of stored intonation descriptions in the device.

18. Device according to claim 13, characterized by the model instrument being arranged to analyse the lexical information regarding syntax at which the model is brought to correspond to one in the language standardized pattern.

19. Device according to claim 18, characterized by the device being arranged to categorize the speech in different dialect categories on the basis of stored intonation descriptions in the device.

20. Device according to claim 13, characterized by the device being arranged to categorize the speech in different dialect categories on the basis of stored intonation descriptions in the device.