Speech Enhancement Technique Research In Low SNR Circumstance

In general, speech is often corrupted acoustically by ambient noise. The result is the degradation of digital voice processor .So a system for speech enhancement is needed strongly to solve the problem. The objective of speech enhancement may be to improve the quality, to increase the intelligibility, to ensure the reliability of digital voice system. So, researching speech enhancement has important value in actual application.This thesis can be divided into two parts: noise's estimation in low snr circumstance and speech enhancement. The latter is the main part of this thesis.Mass speech enhancement arithmetic, such as self-adaptive filter, spectral subtraction, need to estimate noise accurately. Usually we use traditional voice activity detection arithmetic to estimate noise, but those methods's detection correctness is so poor in low snr circumstance to influence estimate noise. In this thesis, we bring forth a new voice activity detection method on the maximum of autocorrelated functions based on speech's cycle. Experiment proves that this method is adapted to estimate noise in low snr circumstance for speech with noise.The second part concentrates on spectral subtraction, mainly buildup spectral subtraction that is current an enhancing technique to dispose broad band noise. This method is flexible, though it cannot synchronously reduce speech distortion and rudimental noise. Contraposing this disadvantage, this thesis bring forward improvement for this method. Firstly, this thesis introduce prior snr, bring...