Speech Enhancement Technology And Its Dsp Multi-channel Real-time Implementation

Speech enhancement is an important branch of speech signal processing. It is also an important pretreatment technology of speech coding, speech recognition and speech synthesis.Among of many algorithms, spectral subtraction method has been one of the most well known techniques for noise reduction because of its simple and effective, but the enhanced speech would be corrupted by a residual noise named musical noise at low SNR. Due to a wide range of adaptability of SNR and effectively restraining musical noise at low SNR, the speech enhancement algorithm based on minimum mean square error has become a research hotspot at the field of speech enhancement.In this paper, firstly, the basic theory of speech enhancement has been introduced. After analysis and research of principles of spectral subtraction and minimum mean square error, the priori SNR and the noise power spectrum estimation of original algorithm based on minimum mean square error have been improved. The improved method can obtain more accurate priori SNR and real-time update the noise power spectrum estimation.Then, multi-band spectral subtraction, extended spectral subtraction, classical minimum mean square error method and the improved method are simulated in a variety of noisy conditions. The results show that the improved method get better performance in the evaluation of SNR and Itakura distance.Finally, the improved method is implemented with standard C language based on DSP algorithm standard. Validation and testing are made on TMS320VC5509A and TMS320C6416T DSP hardware platform. The results show that the module retains the performance and supports real-time multi-channel audio signal processing.