Performance Assessment Of The Denoise In Speex And Its Porting For The ARM Platform

The realistic telecommunication systems are always equipped in various noisy environments, where the quality of the speech would be badly impaired as a result of its noise interference. Noise suppression, one of the most important technologies in speech signal processing, can improve the quality of the speech and the intelligibility of the speech. Numerous effective algorithms have been developed and applied for noise suppression for speech processing, such as speech coding and speech recognition. Most speech processing devices take the ARM embedded processors as their cores, whose preliminary requirements are small volume and low power consumption, as well as low complexity and high efficiency for the program. Efficient noise suppression programs operated on the ARM embedded processors will of vaster market values in the future. Therefore, porting and optimizing the noise suppression program for the ARM platform is gaining more and more significance.This thesis mainly focuses on the porting and optimization of the noise suppression module in the Speex speech codec for the ARM platform. Based on the analysis of the enhancement algorithm in the noise suppression module, a speech quality assessment of its denoising result and the voice quality recovery is implemented. The segment SNR function and the PESQ objective evaluation method that is most relevant to subjective evaluation method are adopted in the assessment. Results of the assessment prove that the noise suppression module performs well on both sides. In view of the wide application of the ARM embedded processor in the communication terminal field, a test of its running-efficiency on the ARM platform after porting the noise suppression module is implemented. And the optimization was carried out because of its low running-efficiency on the ARM platform. Optimization is the focus of this paper. Owing to the complicated implementation of the code optimization, it took a long time and extensive energy to carry out the optimization, in which two kinds of optimizing methods were employed, in terms of C code optimization and ARM assembly optimization. After testing, in premise of the speech quality, the optimized code heightens its running speed in line with the satisfaction of the demand for real-time applications.