The Pretreatment Of Speech Signal Based On Modules

In modern information technology, the processing of the speech signal is becoming more and more important, in which the pretreatment of the speech signal is the necessary part. It will improve the quality of the process of the speech signal. It mainly consists of the automatic level control, adaptive noise suppression, high level compensation and howling suppression, which are the key technology in this paper.Automatic Level Control (ALC) is mainly used to control the strong fluctuation of speech signal. It adjusts the level of the speech signal to the best value timely. ALC mainly contains two parts. One is two classes dynamic VAD, the other is automatic gain control base on RMS.Adaptive noise suppression is used to pick up the relatively pure speech signal from the speech signal with noise. In this paper, there are some improvements based on the spectral subtraction, such as eliminating the "music noise" by mask line, carry out the noise suppression by least control.In advantage of the hearing characteristics, a high level composition based on equal-sound line functions is proposed. This method the speech frames are chose by VAD first, then calculate the parameters of the level composition. At last, compositing the high frequency and low frequency of the speech signal multiplied by the parameters. So the equal sound is obtained.Phase randomized technique is proposed through analyzing how to form the howling. First, put the original signal into polyphase IIR filters to get the sub-band signal; second, the exits of howling is detected by LMS adaptive notch filter; if the results of the detection is howling exiting, restrain it by phase randomized technique.The modules of the pretreatment of the speech signal mentioned above not only can process separately, but also can work together as a whole system. So it will use different modules according to different situations to achieve the different aims.