Research On CS Based Speech Enhancement And Speech Processing Simulation Platform Development

In reality, the speech signal is always inevitable interference by other signals, resulting in degradation of signal quality. It has tremendous stimulative effect to promote people’s production and life to improve the quality and ways of speech communication. How to improve the quality and intelligibility of the speech signal as much as possible is an important part of speech enhancement algorithm.Compressed sensing (CS) is a novel signal sampling and compression theory, which can reconstruct an original signal accurately or with high probability of accuracy by using fewer signal sampling values, where useful information is extracted and retained effectively. CS is applied into speech enhancement processing in this paper firstly, and a new speech enhancement algorithm based on CS is proposed. The algorithm uses the effective extraction and recovery of CS process to realize speech enhancement, raises a new idea for speech enhancement techniques.Double thresholds orthogonal matching pursuit (DTOMP) is proposed in this paper, which is improved on the basis of orthogonal matching pursuit (OMP). Similarity and energy threshold are used to control reconstruction iterative process in DTOMP, which can recover enhanced speech under white noise, color noise and speech-like noise environment.Adaptive quantile noise power estimation algorithm is proposed based on the research in the traditional quantile, used to estimate the noise power of noisy speech, and then set energy threshold of DTOMP algorithm. Improved adaptive quantile method does not require initial value settings, compared to the traditional quantile, the noise power estimate is more accurate.Experiment under white and color noise with different signal to noise ratio (SNR) condition, SNR and log spectrum distortion (LSD) are used to objective evaluation of the proposed algorithms. Experiment results verify the performance of the algorithm proposed in this paper applied to speech enhancement processing is superior in the spectral subtraction and the subspace algorithm.A simulation platform for algorithms simulation of digital speech signal analysis and processing is designed and developed in this paper. In accordance with component object model (COM) specification, using C/C++development language, based on actual command, the main modules of simulator forms are designed and implemented, which includes controller, simulation interface, project management and other important functional blocks. Simulator platform realizes speech signal analysis and processing algorithms for visual editing and debugging, easy to use, flexibility and scalability.