论文部分内容阅读
本文针对传统语音参数对声母等不平稳结构识别中的不足,对语音不平稳部分进行了统计分析,并根据其特性,提出了一系列新的从平稳性研究出发得到的语音参数,并将其应用于一组音长短、动态性强的塞音声母的识别.实验表明,与仅用传统的LPC系数相比,集外测试识别率提高了19个百分点.多元统计分析表明,新参数的加入有利于区分各个声母,并且其作用独立于现行的语音参数.这些参数较好地说明了语音的平稳度,并且计算简单、易于实现.
Aiming at the shortcomings of traditional speech parameters such as initial consonants and other unsteady structures recognition, the unsteady part of speech is analyzed statistically. Based on its characteristics, a series of new speech parameters are derived from the stationary study. Which is applied to the recognition of a consonant of short and dynamic consonant.Experiments show that the recognition rate of extra test is improved by 19 percentage points compared with the traditional LPC coefficient only.Multivariate analysis shows that the addition of new parameters Which is helpful for distinguishing each consonant, and its function is independent of the current voice parameters.These parameters better illustrate the flatness of the voice, and the calculation is simple and easy to implement.