Realization And Optimization Of Speech Enhancement Based On Dm6446

Speech is the most effective and convenient means for the exchange of information. The research on the speech processing technology with modern means makes it effective to generate, transmit, store, access human voice, which is very significant to the social progress. However, the speech is hardly pure due to the various effects in the real world, resulting in great difficulty during the speech processing. A platform with high-performance multimedia processor was exploited to implement the noise and echo cancellation algorithm for speech enhancement, and it could be integrated into a variety of application specific and complex embedded systems for speech processing.The noise reduction and adaptive echo cancellation algorithm for speech enhancement was analyzed, focusing on NLMS algorithm in adaptive echo cancellation and Wiener filtering algorithm in noise reduction, and an improvement to Wiener filtering algorithm was achieved. The algorithms were implemented in the DM6446 embedded platform. After finishing the float-point C code for the improved Wiener filtering algorithm and NLMS algorithm, a fixed-point process was operated. The proposed Wiener filtering algorithm showed a significant improvement comparing to the original one, tested by ears and simulations. The error was less than 0.2% after the fixed-point process, showed by the comparison with the previous value.DaVinci digital SoC system DM6446 was selected as the verification platform. The self-start function was realized by the transplantation of BootLoader and Linux operating system(OS). Montavista Linux2.6.10 OS in ARM core was exploited to control the system, including the set of sampling frequency & bit-width and voice capture & playback. The speech enhancement algorithm was implemented by DSP core. The algorithm optimized aiming at the platform achieved an improvement of nearly 130% in operational performance.