基于最大似然估计的子带语音去噪的研究

对基本的谱减算法进行了改进。利用满足听觉感知模型的滤波器组对语音信号进行处理,然后通过自动跟踪信号的低能量部分的包络来估计背景噪声的特性,最后利用改进的谱减技术对子带语音进行滤波并增强。实验表明,改进的谱减技术具有良好的性能。

The Study of Sub-band Speech Spectral Subtraction Based on ML

Based on sub -band speech processing,this paper improves the basic spectral subtrac-tion.At first,we process the speech with PR filter bank which satisfied a uditory models;and then,we esti-mate the noise statistical characte r by tracking the low -energy envelop e of the signal within frequency band;finally,we filter and enhance the speech signal with the improved s pectral subtraction.Experiment show that the improved speech subtra ction enhance the performance of system.