The five feature parameters, vocal tract area functions, PARCOR coefficients, FET cepstrum, LPC cepstrum and Weighted Likelihood Ratio (WLR), are compared in recognition experiments Using Continuous Dynamic Programming (CDP). The post processing of the CDP which makes a word lattice is used and no syntactical constraints are introduced. The speech materials analyzed are the continuous utterances of 54 principal city names. The highest discrimination rates are obtained with LPC cepstrum (97. 8%) and WLR (97. 4%).
|ジャーナル||Bulletin of the Electrotechnical Laboratory, Tokyo|
|出版ステータス||Published - 1984|
ASJC Scopus subject areas