Speech Enhancement Algorithm And Signal Analysis Research And Development Platform

As the commonest and most direct method for information expression, the speech signal is widely used in many fields. However, during the process of voice communication, the signal will unavoidably be contaminated by noises from the background and transforming medium, electrical noises within the communication devices and even other speakers. Because of these disturbances, the receiving speech signal is not clean and original but the polluted one. In order to extract the underlying clean speech signal from the noise speech signal and reduce the noise disturbance to the minimum, speech enhancement is needed.In general, to eliminate the noise completely is impossible, because the disturbing signals randomly happen. Thus, the main goal of speech enhancement rests with enhancing the speech quality, improve the speech intelligibility and relieve the fatigue feeling for the receiver; and for the speech processing system (recognizer, vocoder, handset, etc.), the destination is to improve the recognition rate and anti-interference of the system.In this paper, the summarization of current typical speech enhancement methods and their characteristics are first introduced. Then, we present the newer multi-taper spectral algorithm and probe the method into many aspects, such as the application of multitaper spectral, the selection of wavelet function and the application of masking effect in speech enhancement. Considering the deficiencies of hard threshold method and soft threshold method, an improved wavelet threshold function is proposed. At the same time, in order to convenience the speech signal analysis and application, the author joins in the development of analysis platform for speech signal. We proposed some particular methods to solve the encountered problems during the period of development, such as multithread, zoom of the waveform, etc., and constructed and designed a framework for the analysis platform. The programming code written in the system can be used for the future application for speech signal.The comparison experiments are further developed within the threes methods, which are spectral subtraction, STSA-MMSE and multitaper spectral algorithm. Experimental results show that the enhancement effect of multitaper spectral algorithm...