Research Of Speech Enhancement Based On Hilbert-Huang Transform

As the developing and application of communication and multi-media techniques, the requirement of speech quality is becoming more and more rigorous. However, speech signal would be interfered by environment noise in various degree in real application environment, which decline the performance of the speech processing system. Speech enhancement is to take the situations like that out of the corner, restraining the noise, improving the quality and make it more comfortable to human hearing, and getting the most pure speech signal out of the noised signal.Based on the result of other's researches, the enhancement of noised speech signal is studied from the perspective of Hilbert-Huang Transform analysis method. A few new solutions for the voice activity detection, the noise estimation and the speech enhancement of noised speech signal are proposed. Aimed in the enhancement of the noised speech signal, the emphases are as follow:1) Based on the advantages of the Hilbert instantaneous energy spectrum in expressing the instantaneous energy of the speech signal, one voice activity detection method based on Hilbert instantaneous energy spectrum is proposed and applied to the speech signal;2) Combing the employ of minima controlled recursive averaging method (MCRA) with the result of voice activity detection to estimate the noise in speech signal;3) Based on the HHT analysis method, improved spectral subtraction as well as a quasi-wavelet soft-threshold method and an instantaneous frequency based signal reconstruction method is proposed to realize the speech enhancement. Take the quasi -wavelet soft-threshold method as the example, based on the coefficient of the power ratio of the speech signal and its' IMF in speech stage, the enhancement is achieved in IMF level. In conclusion, both subjective and impersonal evaluating method are introduced to make the comparison and evaluation of the above three method proposed.