基于独立成分分析的双麦克阵列语音增强算法

提出了利用频域的独立成分分析（Independent components analysis）算法分离语音信号和噪声信号,达到抑制噪声的效果.并且,针对ICA算法在噪声源集中的环境中效果较好,在噪声源分散的环境中性能有所退化的情况,基于时域带噪信号的ICA算法提出频域带噪信号的ICA算法.最后利用最小均方误差估计谱幅度算法（Minimum mean square error）去除残留噪声,达到较好的语音增强效果.通过大量的实验数据测试,文中提出的基于ICA和MMSE短时谱幅度估计的双麦克语音增强算法在不同信噪比（Signal to Noise Ratio）下,都取得了良好的降噪效果.

Speech enhancement using two-microphone based on independent component analysis

The proposed algorithm applies the frequency-domain ICA to cancel noise. The original frequency-domain ICA has shown impressive noise reduction abilities in a directional noise field, while maintaining low speech distortion. However, in a diffused noise filed less significant noise reduction is obtainable. In this contribution this paper proposes two methods for improving the performance in diffused noise field. One extends the basic ICA model to the situation where noise is present. The other uses the MMSE estimate of spectral amplitude as a post processing. An the experimental study, which consists of both objective and subjective evaluation in various noise fields, demonstrates the advantage of the proposed algorithm.