Research On Speech Enhancement Algorithm And Realization Based On DSP

Noise in a complex environment affects the quality of voice communication seriously,and can not convey the semantics correctly. Speech enhancement technology can extract useful speech signal from background noise, restrain and reduce the noise interference, and improve the naturalness and the intelligibility of speech. It has become a key researching point of speech signals processing field in the world.This paper introduces the basic theory of speech signal processing, and the voice activation detection algorithm is discussed in detail which acts as front-end analysis to speech enhancement processing. On the basis of the existing activation detection algorithm, an improving design of activation detection algorithms is proposed in this paper which is based on short-time energy and zero crossing rate double thresholds. The algorithm improves the accuracy of judging the speech endpoint by comparing the energy of speech frame with the characteristics of the short-time energy, and the accuracy of judging the starting point of the speech signal and voiceless consonant and silent period between the cut-off points with zero crossing rates. In view of the problem of strong music noise and serious distortion of noise,this paper puts the improving power spectral subtraction algorithm on the basis of the existing basic power spectral subtraction forward. The algorithm improves the quality of the speech signal greatly by increasing the value of attenuation coefficient and spectrum smoothing coefficient moderately, and by adding the number of index parameters to achieve the best compromise among noise suppression, the residual noise attenuation and speech distortion.Considering the hardware implementation, this article takes the TMS320C6713 DSP chip with the high speed frame processing ability as the control core of the system.Meanwhile, multi-channel buffered serial port is used to realize connect seamless connection with the extended audio interface chip. The interface circuit is designed between the phonetic acquisition/replay module, external memory chip and the DSP chip. So, the complexity of the system design is simplified and the system integrates these functions of high speed voicesignal acquisition, de-noising processing and output. In the aspect of software implementation,these initialization configurations of audio processing chip, multi-channel buffer serial interface and enhanced direct memory access are discussed, and these algorithms of the improving voice activity detection technology and improving power spectral spectrum subtraction technology are realized in DSP.Furthermore, improving speech enhanced algorithms proposed in this paper were simulated in the MATLAB environment and were tested in DSP system designed in this paper using different types of voice signals with noise. Simulation and test results showed that the signal-to-noise ratio of these test voice signals was improved through the algorithm processing, these voices becomes clearer and more intelligibility and the purpose enhancing the useful speech was achieved.