An approach to linear prediction coefficient (LPC) analysis basedon the normalization of vocal-tract length is presented. The approach isof significance for speech recognition of arbitrary speakers. In thisapproach, the ratio of two vocal-tract lengths corresponding to a newspeaker and a reference one is first estimated from the training speechdata of several typical vowels. The LPC parameters normalized on thisratio can then be calculated for any speech data. Compared with previousmethods of speech parameter normalization, this approach does not needto estimate formant frequencies and is simple and reliable in theory.Limited experiments on the recognition of nine Chinese vowels for fourspeakers to indicate that this new approach can achieve 5% to 20%improvements of correct recognition rate
展开▼
