The Research And Implementation Of Speech Enhancement

Speech enhancement is an important branch of Speech signal processing, which is often widely used in speech coding and speech recognition systems as a front-end voice processing. With the continuous development of DSP chips and the improvement of its supporting development tools, real-time speech enhancement has been well implemented, and has been widely used in real environment on different occasions. The aim of the actual speech enhancement is extracting pure original voice from the noisy speech, improving voice quality and eliminating background noise, so listeners willing to accept without feeling fatigue, improving intelligibility, making listener to easy understand.Speech enhancement algorithm needs to estimate the noise spectrum first , which directly affects the accuracy of the final speech enhancement. In this paper , the energy threshold VAD detection and minimum constraints recursive noise power spectrum estimation are described, that both have advantages and disadvantages, but because the later has too many adjustable parameters, the Matlab simulation and real-time DSP implementations in later chapter choose energy threshold VAD as noise spectrum estimation.In the paper, three typical speech enhancement methods of single channel short spectral have been studied, including spectral subtraction, Wiener filtering and minimum mean square error method, and analyzed their strengths and weaknesses. All three enhancement algorithms have been made a modest improvement and simulated by Matlab, the voice source is mixture from street noise and pure voice of Neospeech database in different SNR, and the SNR range is -5 dB ~ 20 dB. subjective and objective evaluation results show the effect of speech enhancement is good.This paper introduces the hardware and software design of speech enhancement system based TMS320C5402 DSP and TLV320AIC23. The interface design between AIC23 and DSP and its associated registers'configuration are mainly introduced ,the peripheral circuits of AIC23 are described briefly. Replay detection shows the hardware working properly. In the software part of the selection, spectral subtraction is selected as speech enhancement algorithm implementing in real-time, and its code is optimized partly. The CCS results show that the optimization to achieve a certain effect.