Speaker differences in speech and speaker recognition
Abstract
The range of ways in which speakers differ is surveyed, with distinctions being drawn on the one hand between physiological and usage differences and on the other hand between those differences stemming from the larynx and those stemming from the vocal tract. Methods of dealing with speaker differences in speaker-independent and speaker-adaptive speech recognition systems are discussed. This is followed by a discussion of the exploitation of speaker differences in speaker recognition systems. The latter discussion is divided into a consideration of speaker verification, in which a speaker is trying to prove his identity, and speaker identification, in which the identity of an unknown speaker has to be discovered and the speaker cannot be expected to cooperate in producing a predetermined phrase. A summary of the present state of the art in dealing with speaker differences together with some guesses about the prospects for practical systems in the near future.
- Publication:
-
In AGARD Speech Process. 8 p (SEE N83-34179 22-32
- Pub Date:
- May 1983
- Bibcode:
- 1983sp...agarR....H
- Keywords:
-
- Frequency Distribution;
- Human Factors Engineering;
- Speech Recognition;
- Automatic Control;
- Error Analysis;
- Languages;
- Markov Chains;
- Phonemes;
- Power Spectra;
- User Requirements;
- Communications and Radar