摘要： 为优化基本减谱法在低信噪比下的性能，必须对噪声进行更为准确的估计。提出一种基于谱熵噪声估计的改进减谱法，通过谱熵值估计每一帧语音的噪声，利用减谱法减去噪声以达到降噪的目的。实验结果表明，该方法与传统消噪方法相比，在不同噪声环境下均能取得较好的去噪效果。

关键词: 谱熵, 减谱法, 语音识别

Abstract: In order to improve the performance of traditional spectral subtraction in low SNR environment, more accurate noise estimation is needed. This paper proposes a novel spectral subtraction based on noise estimation of spectral entropy. It estimates the noise of each frame according to its spectral entropy, and subtractes it to enhance the frame. Experimental results show that in comparison with other classical methods, the proposed method can obtain better performance in different noisy environment.

Key words: spectral entropy, spectral subtraction method, speech recognition

中图分类号: 

• TP391