基于听觉掩蔽效应的改进MFCC特征提取算法

目前，关于语音识别的研究尚处在实验室环境中，而实际的语音总是与噪声和干扰并存。人类能够在信噪比很低甚至在有干扰声音存在的环境中正确识剐语音主要是依靠人的双耳输入作用，本文就模仿人耳的听觉掩蔽效应来掩蔽噪声信号，提出了一种MFCC（Mel频率倒谱系数）改进提取算法。该算法能更好地减少噪声信号对纯净语音信号的影响，从而提高语音信号的识别率。实验表明改进后的算法相对于传统的MFCC提取算法大约有4．43％～8．42％的相对性能提升。

Improved MFCC Feature Extraction Algorithm Based on Hearing Masking Effect

The research on speech recognition is still in the laboratory stage, and the actual speech always exists in noises. Human can exactly identify the speech which has low signal-noise-ratio even with jamming voice. This mainly depends on the input function of both ears. In this paper, an improved extraction algorithm for Mel frequency cepstral coefficient （ MFCC ） is proposed, which is based on the heating masking effect. This algorithm can decrease the influence of noise signal to the speech signal, and improves the recognition rate. Experiment shows that the improved algorithm has about 4.43% -8.42% relative performance improvements.