一种改进型谱减法的电子商务语音噪声去噪方法研究

Research on An Improved Spectral Subtraction Method for Electronic Commerce Speech Noise Reduction

文章针对语音信号经过传统谱减算法增强后仍残留"音乐噪声"的问题,提出一种改进的加权幅度谱估计多带谱减法。首先对带噪语音信号进行加权幅度谱估计预处理,减少谱估计的波动,其次将带噪语音信号按照频率划分成不同的频带,最后通过语谱图分析增强后的语音效果。实验结果表明,这种改进型的谱减法具有更强的抑噪效果。

An improved weighted amplitude spectrum estimation multi-band spectral subtraction method is proposed to solve the problem of musical noise remaining in speech signal after enhancement by traditional spectral subtraction algorithm.Firstly,the noisy speech signal is preprocessed by weighted amplitude spectrum estimation to reduce the fluctuation of spectrum estimation.Secondly,the noisy speech signal is divided into different frequency bands according to frequency.Finally,the speech effect after enhancement is analyzed by spectrogram.The experimental results show that the improved spectral subtraction method has stronger noise reduction effect.