Research Of Speech Enhancement Algorithm Based On Microphone Array In Indoor Environments

In indoor environments, the speech signal received by microphone is influenced by environment noise(e.g. diffuse noise), directional noise and reverberation. Among the existing microphone array speech enhancement algorithms, the Convolutive Transfer Function Generalized Sidelobe Canceler(CTF-GSC) can decrease the directional noise effectively while unaffected by reverberation. However, the CTF-GSC cannot be applied to an enclosed environment. The post filtering algorithm based on single channel wiener filter can decrease diffused noise largely while keep a minimum signal distortion. This algorithm has been applied to many microphone array speech enhancement systems. Aiming at the noises and reverberation, this paper gave emphasis to study the post filtering algorithm based on single channel wiener filter and CTF-GSC algorithm. This topic is undertook by Shenzhen technical innovation projection—the array signal speech processing technique research based on smart TV, project number: CXZZ20130517113418268. The main work done in this paper is as following:1) Study on post filtering algorithm. The traditional wiener algorithm has a serious speech distortion when it towards low signal to noise ratio(SNR). Aiming at this disadvantage, an improved algorithm was proposed. This paper used the speech signal collected by LabVIEW audio collection system to do simulation experiment, the results showed that, with sacrificing very little speech SNR, this algorithm can receive lower the speech distortion degree and make the enhanced speech rise to a more natural level. In the white noise field, although the speech enhancement based on improved Wiener filter improve 0.02 dB segmental SNR improvement than the traditional Wiener filter on average, the degree of distortion can reduce 3.15% than the traditional Wiener filter.2) Aiming at the direct noise in strong reverberation environments, this paper proposed subband CTF-GSC algorithm which uses subband adaptive filtering on the adaptive noise canceler model. The simulation results showed that the subband CTF-GSC algorithm can improve the noise reduction and keep low computational complexity, consequently, increasing the convergence speed.3) In an enclosed environment, considering the influence of diffused noise, directional noise and reverberation can caused on the speech enhanced system, this paper combined the improved post filtering algorithm based on single channel wiener filtering with subband CTF-GSC algorithm to get a new improved CTF-GSC algorithm. The simulation results showed that this algorithm not only can suppress the directional noise effectively in s strong reverberation environment, but also can reduce the diffused noise effectively.