A method to select word-candidates for speaker independent word recognition is described. LEP cepstral coefficients are used as feature vectors, and vector quantization method is used to quantize and compress the features in each analysis frame to obtain speaker independence. Experiments were carried out for 100 city names uttered by 10 speakers with good results.
|ジャーナル||Bulletin of Centre for Informatics (Waseda University)|
|出版ステータス||Published - 1986 3月 1|
ASJC Scopus subject areas