A method is discussed for characterizing speaker individuality that could be used for speaker adaptation for speaker-independent phoneme recognition. To represent the individual traits of speakers, a speaker space is organized using distributions in a vector-quantized space. This space includes 200 speakers, each shown as a point in it. The experimental results for 200-speaker identification with 10 input words shows performance rates of 82% in first-order and 90% in the second-order case.
|Title of host publication||Proceedings - International Conference on Pattern Recognition|
|Number of pages||3|
|Publication status||Published - 1986 Dec 1|
|Name||Proceedings - International Conference on Pattern Recognition|
ASJC Scopus subject areas
- Computer Vision and Pattern Recognition