Studies On Speech Enhancement Based On Analyzing Multi-band

Speech signal is often accompanied by the background noise which causes many negative affects, such as polluted speech makes listeners feel tired and it degrades the performance of speech signal process. Therefore, speech enhancement is an important technology of the speech signal process.In real world, noise is mostly colored and does not affect the speech signal uniformly over the entire speech spectrum. To reduce the"musical noise"produced by basic spectral subtraction, a multi-band spectral subtraction method for enhancing speech corrupted by white and colored noise is studied. The variation of signal-to-noise rate is taken into account to confirm the subtraction factor in each frequency band. And we analyzed the improvement of speech quality after speech enhancement by linear and non-linear multi-band bark scale frequency spacing approaches. Experimental results show that both methods can improve the speech quality while non-linear multi-band bark scale frequency spacing approaches is better than linear frequency spacing approaches.Then human auditory masking is studied where its characteristic is combined with the multi-band spectral subtraction method. It efficiently reduces the musical noise and improves the comfort of the human auditory.Besides, the speech pause detection and noise spectrum estimation are researched for accurate estimation the noise statistical characteristic. The method named speech pause detection for noise spectrum estimation by tracking band power is used to do the speech pause detection and improve the noise estimation.Finally,a speech enhancement system based on multi-band spectrum subtraction method is designed to process the speech corrupted by random white and factory noise in different SNR . It proved that the method can largely reduce musical noise and improve speech quality.