时频分布不同特性进行语音分类

Classification of Speech Using the Different Properties of the Time-Frequency Distributions

时频分布在非平稳信号的分析和处理中具有重要地位,它能够直观、合理的描述信号在时间-频率域上的能量分布。语音信号分类是语音识别、说话人识别、语种辨识和语音合成的一个重要基础,而信号表示的方式和距离测度的选择,对分类性能影响很大。该文正是利用时频分布的特性,对其核参数进行优化,并结合距离测度,完成了独立音标的说话人辨认,获得了较高的准确率,误判率仅为0.99%,具有较好的应用结果。

Time-frequency distributions have an important position in the analysis and processing of the nonstationary signals,which can visually and reasonably describe the energy, distribution of the signal in time-frequency domain. Classification of speech signals is an important base of the speech recognition,speaker recognition,language identification and speech synthesis,and representation of signals and the choice of distance measures dramatically the performance of the classification of signals.In this paper,we have accomplished the speaker identification who spoke individual phonetic symbol with a good result,the 2.48% probability of classification error.The method is just using the properties of time-frequency distributions,optimizing the parameters of kernel and combining time-frequency distributions with distance measures.