Abstract: In order to find a speech de-noising method suitable for most of the environment, the paper presents a de-noising technique based on wavelet-spectrum analysis. The features of this method include：the multi-scale analysis for noisy speech can be done by using multi-resolution of wavelet transformation, and the speech and noise can be distinguished by using characteristics of self-correlation of speech spectrogram, as well as the residual noise of speech segment can be cleared by using continuous monitoring of points. Experiments show that the de-noising method of wavelet spectrum analysis can remove a variety of wide-band noise in different environments.

Key words: de-noising technique, wavelet decomposition, speech spectrogram, self-correlation, wide-band noise

摘要： 由于不同环境下噪声特性不同，多种环境下的语音去噪成为研究难点。提出一种基于小波语谱图分析的去噪技术。该方法的特点在于：利用小波变换的多分辨性对带噪语音进行多尺度分析，利用语谱图列自相关函数的特性划分语音段和噪声段，利用点连续检测法去除语音段残留的噪声。实验显示，小波语谱图分析去噪法对多种环境下的宽带噪声，抑制效果显著。

关键词: 语音去噪, 小波变换, 语谱图, 自相关函数, 宽带噪声