基于改进基音跟踪算法的单通道语音分离

Monaural Speech Segregation Based on Improved Pitch Tracking Method

基于计算听觉场景分析(Computational Auditory Scene Analysis,CASA)的语音分离系统通过模拟人耳的听觉感知系统对混合信号进行处理并分离出感兴趣的目标语音,近年来得到了很大的发展。如何在干扰噪声存在的情况下进行正确的基音提取跟踪一直是CASA系统研究的重点。提出了一种基于目标语音源的改进基音跟踪算法。该算法通过对目标源估计和基音检测两个步骤的反复迭代计算,得到最终的基音轨迹。通过在不同噪声干扰条件下与传统基音跟踪算法对比的实验结果证明,该算法能够有效地抑制噪声,提高输出语音的信噪比和语音质量。

By means of the feature that humans can distinguish and track speech signal of interest under various noisy environments, the speech segregation system based on computational auditory scene analysis （CASA） has obtained considerable development in recent years. How to correctly pitch detection in noisy environment has been a challenge in CASA system. Hence, in this paper, an improved pitch tracking algorithm based on target source is proposed. By estimating the target units and detecting the pitch periods iteratively, the proposed algorithm obtains pitch tracks. By comparing with conventional pitch tracking method under various interferences, it is shown that the proposed algorithm can effectively suppress the interferences and improve the average output SNR and the quality of segregated speech.