基于声音定位和听觉掩蔽效应的语音分离研究

Speech Separation Based on Sound Localization and Auditory Masking Effect

人耳具有在嘈杂环境中将感兴趣的语言信息提取出来的能力 ,而双耳听觉特性有助于这种能力的加强 .据此本文提出了一种基于声音定位和听觉掩蔽效应的混叠语音分离方法 .根据声音到达双耳的时间差和强度差在时频域内确定相应的掩蔽系数 ,该系数是二值的 ,以直接去除干扰信号 ,保留有用信号并达到语音分离的目的 .实验表明 ,本文提出的方法是有效的 .该方法不仅适用于混叠语音为浊音情形 ,对清音的情况同样适用 ,因而比基于基音提取的语音分离方法的适用范围更广 .

Human has the ability to attend to a single interested speech in a noised condition and this ability can be improved in the presence of binaural cues. In this paper a speech separation method is presented based on sound localization and auditory masking effect. By two important parameters-the interaural time differences (ITD) and interaural intensity differences (IID)-we estimate the binary masking coefficients in corresponding time-frequency regions. The coefficients are helpful of speech separation by holding interested signal and reducing noise signal. Experiments indicate that the approach described here is efficient not only for voiced speech but also for unvoiced speech and it has more extensive applications than pitch-based speech separation algorithms.