基于计算听觉场景分析的混合语音信号分离算法研究

Research on speech separation based on computational auditory scene analysis

人耳听觉系统能够在强噪声的环境下区分出自己感兴趣的语音,基于计算听觉场景分析(CASA)的基本原理,其重点和难点是找到合适的声音分离线索,完成目标语音信号和噪声信号的分离。针对单通道浊语音分离的问题,提出了一种以基音为线索的浊语音信号分离算法。在白噪声、鸡尾酒会噪声等六种噪声干扰条件下,通过仿真实验结果表明,相比于传统的谱减法,语音分离算法的输出信噪比平均提高了7.47 d B,并有效抑制了干扰噪声,改善了分离效果。

The human auditory system shows a remarkable capacity for speech segregation. Based on the principle of computational auditory scene analysis（ CASA）,the vital task was to find the right auditory cues to separate target voice. For monaural speech segregation,this paper proposed voiced speech separation algorithm based on pitch clues. Under the condition of six different noises,such as white noise,cocktail party noise,the experimental results show that the average target speech SNR is improved by 7. 47 d B compared with the traditional spectral subtraction,this proposed algorithm also effectively restrained the noise and improved the performance of voiced separation.