Study On Combined De-noising Speech System Base On DSP

The primary purpose of speech noise reduction is to extract pure speech from speech signal with noise. It is an effective way to solve the pollution from noise. Because of complex and non-stational of speech, especially unvoiced speech has no clear time and frequency character, which is very similar to white noise, classic noise reduction method is unsatisfactory. And especially in broadband case noise reduction effect is not very clear.In this paper, firstly we introduce the basic principle of genetic algorithm, and detailedly describe the binary system genetic algorithm and its concrete operating scheme. According to the characteristics of FIR digital filter, we put forward optimization design of FIR digital filter with an improved zooming genetic algorithm, and use a practical example to prove it has the advantages of fast convergence speed and the whole solution quality.Secondly we research the applications of wavelet de-noising in speech signal processing. We decompose speech signal by frame using db6 wavelet at 4 layers, calculate the energy mean and variance at all layers, and distinguish speech from noise by energy variance. At the same time we discriminate voiced and unvoiced sounds by energy mean. Then we use soft-and-hard-threshold tradeoff method to process determined signal, remove the noise by wavelet threshold de-noising. At last we reconstruct wavelet coefficients and achieve the de-noising speech signal.On the basis of above, we design a combined de-noising speech system based on FIR and wavelet de-noising. We filter out the speech frequency outband signal with FIR digital filter, and process speech signal with endpoint detection, voiced/unvoiced determination, threshold de-noising. At last we design the combined de-noising speech system based on SEED-DEC6416, and realize stand-alone. The results show that the combined de-noising method has a better effect and a practical application value.