Patent
1996-01-11
1997-12-02
Tung, Kee M.
395 216, 395 264, G10L 506
Patent
active
056945205
DESCRIPTION:
BRIEF SUMMARY
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to artificial recognition of different dialects in a language.
2. Discussion of the Background
In speech recognition it is previously known to use speech recognition equipment of different kinds. In classical systems the speech recognition equipment is trained to recognize speech from a large number of persons. The information which is at this achieved is after that used for interpretation of speech coming in to the equipment. The speech recognition equipment is at this trained to recognize speech that follows practised dialects. Dialectal variations in speech outside the practised dialects is with this type of speech recognition equipments not generally interpretable.
In languages with tone word accents and tone language the intonation constitutes a very important part in the understanding of the language. In the previously known technology no consideration has been taken to these considerations. The consequence of this is that the interpretation of words and phrases have been misinterpreted at artificial speech recognition. To the extent speech recognition equipments have been constructed to manage dialects in a speech, these euipments have been especially constructed for the dialect in question.
In the future speech recognition equipments will, to an ever increasing extent, be used in different connections. The speech recognition equipments shall in this connection be capable to recognize different dialects in a language. The dialectal variations in a language have been difficult to describe in one for the machine useful way. The artificial understanding of speech has not in this connection given a satisfying result. Further, there is a wish to find methods which are generally applicable to different languages.
The above mentioned problems have implied that artificial interpretation of speech have been difficult or impossible to perform due to dialectal variations. General methods are therefore of greatest importance.
Beside the pure technical problems to interprete a speech there are strong wishes that the speech shall be possible to interprete in order to control different types of equipments and services in for instance a telecommunication network.
The present invention is intended to solve above mentioned problems.
SUMMARY OF THE INVENTION
The present invention firstly relates to a method to, out of a given speech, recognize dialectal variations in a language. For this purpose a speech recognition equipment is adapted to recognize different dialects in the language. From the speech a fundamental tone curve is extracted and its maximum and minimum values are identified. Out of the speech is further a speech recognition performed from which a model of the speech is created by means of lexicon and syntax analysis. The achieved model is given a standard intonation. The maximum and minimum values of the fundamental tone curves of the speech and the model respectively is compared with each other. A time difference between the maximum and minimum value occurrances in the fundamental tone curve of the speech and the fundamental tone curve of the model respectively is obtained.
This time difference has an effect on the model. The model will at this be adapted to the intonation of the speech. In this way a model of the speech is obtained which, regarding dialect, corresponds to the incoming speech. In this way an improved possibility is achieved to interpret a given speech.
In a further development of the invention a reference is used for determining the time difference at which preferably a CV-limlit is used. Further, the outline of the fundamental tone is based on lexical and syntactic information. In the lexical information is included information about orthography and phonetic transcription. The transcription includes lexical abstract accent information type stressed syllable, toned word accents of type accent 1 and accent 2, and the location of secondary accent, i.e. information given for instance in dictionaries.
The
REFERENCES:
patent: 5581655 (1996-12-01), Cohen et al.
Speech Communication, vol. 15, No. 3-4, pp. 169-186, Dec. 1994, P. Taylor, "The Rise/Fall/Connection Model of Intonation".
European Conference on Speech Communication and Technology, pp. 38-44, Sep. 1989, R. Collier, "Intonation Analysis: The Perception of Speech Melody in Relation to Acoustics and Production".
Patent Abstracts of Japan, vol. 17, No. 707, JP-A-5-241596, Sep. 21, 1993.
Proceedings of the International Conference on Acoustics, vol. 2, pp. 773-776, 1990, J.W. Butzberger, Jr., et al., "Isolated Word Intonation Recognition Using Hidden Markov Models".
Telia AB
Tung Kee M.
LandOfFree
Method and device for speech recognition does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and device for speech recognition, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and device for speech recognition will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-808469