Research And Implementation Reduction Algorithm Based On The Spectrum Of The Masking Effect Of The Human Ear

In recent years, along with the Internet of things'proposed and raised, the importance of speech in communication and signal processing field is more significant. This paper mainly research a modified spectral subtraction algorithm based on the masking property in low signal-to-noise ratio and non-stationary noise. Among focus on single channel spectral subtraction algorithm, and preliminary research the denoising and interference's removal in two channel spectral subtraction algorithm. This paper briefly introduced the thoughts and structures of two double channel spectral subtraction algorithms, and proposed some views after simulations and results'analysis. In comparison with Virag's algorithm, the modification of proposed method in paper is made from three aspects. Firstly, after research, contrast and analysis of two classical noise estimation algorithms(OSMS and MCRA algorithm),the VAD(Voice Activity Detection) is substituted by MCRA(Minima Controlled Recursive Averaging) algorithm to estimate the background noise; Secondly, a modified form of parametric spectral subtraction is employed, the over-subtraction factor is multiplied by a positive constant, which will provide more attenuation to reduce the residual noise; Thirdly, before the input signal implement the high-pass pre-emphasis, do the high-pass filter to remove the low-frequency noise first. Finally, objective and subjective evaluation of the modified algorithm is performed with several noise types, and the results confirm the proposed algorithm's performance is more superior in low signal-to-noise ratio and non-stationary noise, which provide more noise attenuation, more speech retention, and less musical noise, and more pleasant to a modern human listener.