Speech Separation Technology And Application Studiesbased On CASA

With the popularity of intelligent electronic products over the past several years, speechtechnology, especially Automatic Speech Recognition ushered in a new wave of research.While speech signals are usually corrupted by noise degrading speech quality andintelligibility in real environments that application and popularization of speech technologycannot work well.Usually Noise Reduction was used to balance the problem. Although developed a lot,But most of them are based on statistical characteristics of speech, and the speech need tomeet certain constraints, there still exists great challenge. Auditory system of human beingsgreatly exceeds existing acoustic signal handling in perceiving speech signals, so researchersstudy the auditory system hoping to simulate its speech signal perceptual processing. CASA(Computational Auditory Scene Analysis) is a typical algorithm. Researchers develop somespeech noise reduction system with CASA, in which Hu-Wang system stands out. Thesesystems can extract target speech from the mixed signal.In this paper, we show a speech segregation system based on CASA, and proposedsome improvement for Hu-Wang system. Our work includes:1. We proposed a segmentation algorithm using the ratio of high frequency energy andlow frequency energy in a frame to constraint combined cues. Speech signals do not go withhigh energy in high frequency range that reduce the segment result of Time-frequency unit.So tends to improve the merging results of Time-frequency unit during segmentation stage,a new segmentation algorithm based on ratio of every high frequency energy and lowfrequency energy for constraining combined cues is proposed. 2.OQAS(Objective Quality Assessment of Speech) and CASA are mixed in this paper.The evaluation of speech separation usually goes with SNR. While the improvement of SNRdo not means the relevant improvement for the human’s perception effectiveness of speechquality. So we try to mix OQAS and CASA to improve the sound perception with theadvancing of separation signal-noise-ratio gain. Experiment shows that the method canimprove the speech quality of separation.3.The design and realization of improved speech separation system based on CASA.This part blends the two former improved method into the relevant step of speech separationin order to design a complete speech separation system for target speech separation, speechnoise reduction and speech quality improvement.