语音倒谱特征的研究  

Research on Speech Cepstral Features

作　　者：王让定[1,2] 柴佩琪[1] 

机构地区：[1]同济大学人工智能研究室 [2]宁波大学信息科学与工程学院,宁波315211

出　　处：《计算机工程》2003年第13期31-33,共3页Computer Engineering

摘　　要：语音倒谱特征是语音识别中最常用的特征参数，它表征了人类的听觉特征。该文在研究基于线性预测倒谱和非线性MEL刻度倒谱特征的基础上，研究了LPCC和MFCC参数提取的算法原理及提取算法，提出了一级、二级差分倒谱特征参数的提取算法。识别实验验证了MFCC参数的鲁棒性优于LPCC参数。The speech cepstral features are very important parameter in ASR, which symbolizes the property of HAS(human auditory system). The paper discusses cepstral features based on linear predictive and non-linear MEL scale, and presents LPCC and MFCC's extraction algorithm and differential coefficient and accelerated coefficient are also presented. The recognition experiment results show that MFCC's robustness is prior to LPCCs.

关 键 词：LPCC MFCC 特征提取 语音倒谱特征 语音识别 

分 类 号：TN912.3[电子电信—通信与信息系统]