基于预处理VAD和自适应KLT的语音增强算法

Adaptive KLT speech enhancement algorithm with preprocessing VAD

针对加性有色噪声干扰,提出了一种单通道输入基于信号子空间的语音增强算法。算法中使用自适应的方法跟踪KLT(Karhunen-LoeveTransform)阵。运用一种近似模型来表述有色噪声的特性,并基于噪声平稳的假设,通过采用预处理技术的语音活动性检测(VAD:VoiceActivityDetection)单元获取噪声样本,用于下一语音帧中噪声特性的估计和增强处理。实验表明,算法对于有色噪声干扰下的语音信号有较好的增强效果,并且性能优于改进减谱法。

A kind of single input channel speech enhancement algorithm based on signal subspace is studied for enhancement of speech degraded by colored additi ve noise. An adaptive KLT(Karhunen-Loeve Transform)matrix is applied in the a l gorithm. The approximate model of colored noise is made. In the case of steady n oise, preprocessing VAD (Voice Activity Detection) algorithm is used to obtain noise sample. Then it is used to evaluate the character of noise and to accompl ish denoising processing. Simulation results show that this algorithm demonstrat es better performance than modified spectrum subtraction algorithm both in the e nvironment of white or colored noise.