基于波束形成的谱相减语音增强

Speech Enhancement Based on Beamforming for Spectral Subtraction

为降低谱相减算法产生的音乐噪声,并提高语音增强效果,本文在深入研究波束形成技术和谱相减算法的基础上,提出波束形成器后级联谱相减的语音增强处理方法,并分析、验证了这种结构的可行性。对真实环境下的语音和噪声数据的处理结果显示,该级联结构的语音增强系统可显著降低背景噪声,语音失真小,并易于实时实现,信噪比增益达16.5 dB。

A scheme of Frost beamformer followed by the spectral subtraction is proposed based on the beamforming technology and the spectral subtraction algorithm. Then the spectral subtraction can be used in cascading the beamformer. Analysis results show that the structure is reasonable and feasible; Using speech and noise signals recorded in real world the proposed speech enhancement structure can reduce the effects of the noise and the reverberation. The enhanced signal has an almost inaudible musical noise and SNR gain can reach above 16.5 dB.