摘要
压缩感知技术是目前信号处理领域的研究热点。文中针对压缩感知技术实际应用于语音领域时压缩率不高的问题做了研究。首先介绍了压缩感知技术的理论,接着提出了适合建模的观测矩阵的选用标准。在行阶梯观测矩阵下,本文提出了观测序列的一种新模型——基于匹配追踪算法的正弦字典模型,该模型采用稀疏分解的方法对观测序列进行建模。仿真结果表明该建模方法有效地进一步压缩了观测序列,而不明显降低恢复语音的质量。
Compressed sensing (CS) has been a hot spot for research in signal processing. But the com- pression rate of speech CS processing is far below the theoretical value, So some researches are done on this problem. Firstly, the basic theory of CS is introduced. Then, the choice criterion of the measurement matrix for measurements modeling is presented. Finally, in the row echelon measurement matrix, a new sinusoidal dictionary model is established based on matching pursuit (MP) technology, the model uses the sparse decomposition method to model the measurements. Simulation results show that the modelling method can effectively reduce the measurement sequences without significant damage to the recovered speech.
出处
《南京邮电大学学报(自然科学版)》
北大核心
2014年第2期27-31,共5页
Journal of Nanjing University of Posts and Telecommunications:Natural Science Edition
基金
国家自然科学基金(61271335)资助项目