Research And Implementation Of Speech Signal Denoising Algorithm On Communication Device

Speech noise is our life of common problems, and voice noise will degrade in thepresence ofspeech intelligibility and degrade the performance of speech recognition. It isnecessary to reduceor remove the noise from the speech signal, the speech signal ispure reduction. Because a lot ofsoftware algorithm due to the complexity of thealgorithm to the embedded system for noise reduction, and the existing embedded system byusing the method of most of the hardware or the external voice processing chip to process, itis necessary according to the real-time processed by software to improve voice quality.This article will with the characteristics of the wavelet algorithm saves the hardwareresource is given priority to, and the advantages of multiple threshold filter to improve theshortcoming of wavelet in the processing to achieve can be applied to the hardware resourcesis not a lot of hardware, to improve the robustness of speech. Adopted at present commonlyused8KHz sampling frequency, a16-bit linear PCM signal as the input signal, with160sampling points as a frame for the processing of the wavelet.Wavelet transform is a local analysis of time domain and frequency domain of signalthrough scale and balance gradually thinning, finally get the high frequency and lowfrequency segment, can be to get signal time-frequency signal analysis. Noise distribution ofthe original signal in the signal source, and through the voice signal after waveletdecomposition of the wavelet transform system than the wavelet transform coefficient ofnoise, this system uses Daubechies wavelet basis, improvements on multi-layerdecomposition, so as to get fit in with the speech processing decomposition andreconstruction, after decomposition by finding a suitable threshold for the separation of signaland noise, if the threshold is larger, it will lead to useful speech signal has been removed,whereas if processes is not good signal to remove noise reach effect, this thesis by softthreshold processing to achieve a more effective to deal with the noise. Reached a good state.Under the above theory, the existing voice filtering to increase voice part to achieve underhigh noise can hear high quality, make life more perfect.