基于ERB尺度划分的多子带语声信号抗噪谱减算法

A multi-band anti-noise spectral subtraction algorithm using ERB scale

为了研究心理声学在语声增强方面的应用,本文提出了一种基于等效矩阵带宽(ERB)尺度划分的多子带语声信号抗噪谱减算法。此算法根据ERB尺度将带噪信号的频谱划分成多个子带,然后再根据每个子带的分段信噪比以及心理声学掩蔽原则分别计算每个子带的谱减参数,最后在每个子带中分别进行谱减算法处理。实验结果表明,应用新算法所获得的语声增强结果在信噪比、IS失真以及PESQ方面均优于之前提出的多子带语声信号抗噪谱减算法。

This paper addresses a multi-band spectral subtraction algorithm based on equivalent rectangular bandwidth(ERB)scale for applying psychoacoustics to speech enhancement.In the proposed algorithm,the whole spectrum of noisy speech is divided into multiple bands based on ERB scale.The subtraction parameters are then calculated according to the segment SNR of each band and psychoacoustics criteria.Finally,spectral subtraction with different subtraction parameters is executed in each band.The measurements of SNR improvement,IS distortion and PESQ show that the proposed algorithm outperforms the previous speech enhancement algorithms.