基于LPC倒谱的语音特征参数提取

摘要

语音识别指利用计算机识别语音信号所表达的内容，其目的是要准确地理解语音所蕴含的含义。本文着重研究了语音识别实现过程的特征提取，针对特征提取的多种方法，选用LPC倒谱系数作为特征参数提取，较彻底地去除了语音信号产生过程的激励信息，主要反映了声道模型，而且只需十几个倒谱系数就较好地描述了语音的共振峰特性。通过对语音信号进行预加重、分帧、加窗、自相关分析，而后提取出LPC倒谱系数。根据流程编写VC程序，对语音信号进行分析处理，去除对语音识别无关紧要的冗余信息，从而获得用于语音识别的重要信息。%The speech recognition adopts the computer technology to recognize the contents of the speech signal, its purpose is to comprehend the meaning of speech accurately. The paper focuses on the feature extracting of speech recognition; aims at various feature draw methods, it selects LPCC as the extracted feature parameter to exclude thoroughly encourage information which produced in the speech signal creating process. It also represents mainly the sound track model, only more than a dozen LPCCs are needed for a better de- scription of the resonance peak property of speech. Through pre-aggravating, dividing frames, adding windows and self-correlated analy- zing,the LPCCs could be extracted. Program according to process in VC is analyzed, the speech signal is handled, and the insignificant and redundancy information for speech recognition is excluded, then gets the important information which could be.used in speech recognition.