Research On Algorithm Of Speech Compression And Recovery Based On Compressed Sensing

In order to recover the original signal, sampling rate must be greater than the signal morethan twice the highest frequency based on the Shannon sampling theorem, although it isachieved signal acquisition, compression and recovery, but the dramatic increasing incollection and frequency of data is making the current communication systems hard tosupport. Candès et al invented compressed sensing theory, which is well solved the problem.Using compressed sensing theory, signals which are sparse can be compressed fromhigh-level matrix to low-level with linear projection, collecting and compressionsimultaneously worked in this process, in the end, it could accurately reconstruct the originalsignal. Compressed sensing is out of traditional thinking, which will be bound to change thesignal processing in future.In the beginning, the paper reviews the research status of compressed sensing and speechsignal processing depending on it, which introduces the mathematical model of compressedsensing, around signal sparse, designing the measurement matrix and selection ofreconstruction algorithm three key technologies are compared, analyzed restricted property,and displayed differences and connections between compressed sensing and Shannon.Secondly, this paper summarizes the main process of traditional signal processing and thecharacteristics of speech signal. Depending on good compressibility of voice, compressedsensing applied in speech signal compression and storage is an important research direction.Speech signal has good sparse pin the DCT sparse basis, in the final analysis of theexperiment, selecting a speech signal sampling rate of22.05K, the OMP algorithm and BPalgorithm are reconstructed the speech signal respectively, which are analysed subjectivelyand objectively. Finally, it is concluded that the speech signal compression ratio and lengthof frame are both impact on the quality of the reconstructed signal;(2) Using the BPalgorithm reconstructed the speech of quality is better than the OMP algorithm’s, while thedisadvantage of BP is needed more time to recover.In the end, this paper is displayed the shortcomings of sparse groups, observation matrixand reconstruction algorithm of ordinary compressed sensing, and proposed that the adaptivealgorithm is joined in compressed sensing, combined with redundant dictionary KSVDadaptive algorithm, adaptive observation matrix and SAMP reconstruction algorithm, thispaper discovers adaptive compression sensing concept, in the meanwhile, which describes theimplementation process, simulated and compared to verify the concept, what is more, KSVDhas better sparse characteristic according to the number of observations for each frame ofspeech energy, adaptively distributing observation number, and significantly improves thequality of the reconstructed speech, SAMP reduces recovery time of reconstructed signal.Finally, it is simulated and analysed objectively and subjectively adaptive compressed sensing, compared to ordinary compressed sensing, which has characteristics of goodreconstruction quality and few running time etc., thus verify the feasibility of adaptivecompressed sensing.