谱相减技术中“音乐噪声”的抑制

The Suppression of “Musical Noise” in Speech Enhancement by Using Spectral Subtraction

针对应用于单通道语音增强的谱相减技术中存在的“音乐噪声”问题，提出了一种独立的抑制方案。利用Ｕ／Ｖ判决算法以及功率因子α，加权减因子Ｋ的选择较为有效地抑制数字语音系统的前端强输入噪声影响，使处理后的语音信噪比（ＳＮＲ）有明显提高，质量和可懂度也均有明显改善。在不同的噪声特性和信噪比条件下对方案的性能进行的仿真实验测试表明：该算法适用于极低信噪比及缓慢变化的非平稳噪声背景且无须知道噪声的先验知识。

A stand alone noise suppression scheme is presented for reducing “Musical” noise in enhanced speech by using the signal channel spectral subtraction algorithm.The presented method comes from the incorporation of the optimization of voiced/unvoiced decision,power exponent α and weighted subtraction term K .The method effectively suppresses the front end noise of high level in the digital speech processing system,greatly improves signal to noise ration (SNR),speech quality and intelligibility.Its performance has been tested under various noise characteristics and signal to noise ratios with simulation.These tests show that the proposed method also retains the efficiency under very low SNR and slowly time varying nonstationary noise conditions,even if no prior information is given about the noise characteristics or noise level.