摘要
利用语音小波高频系数的稀疏性和压缩感知原理,提出一种新的基于压缩感知的低速率语音编码方案,其中小波高频系数的压缩感知重构分别采用l1范数优化方案及码本预测方案进行,前者对大幅度样值重构效果较好,且不仅适用于语音,也适用于音乐信号,具有传统的线性预测编码方法无法比拟的优势,后者对稀疏系数位置的估计较好,且不需要采用压缩感知重构常用的基追踪算法或匹配追踪算法,从而减少了计算量。两种方法的联合使用能发挥各自的优势,使得重构语音的音质进一步改善。
Utilizing the sparsity of high frequency wavelet transform coefficients of speech signal and theory of compressed sensing, a new low bit rate speech coding scheme based on compressed sensing is proposed. The reconstruction of high frequency wavelet transform coefficients is achieved by li normal optimization and codebook prediction reconstruction respectively, l1 reconstruction has good effect for large coefficients and suits for both speech and music, with which traditional linear prediction coding cannot compare. Codebook prediction reconstruction has good effect for the location of sparse coefficients and reduces the amount of calculation due to not using basis pursuit or matching pursuit. The combination of these two reconstruction methods can bring the advantages of both methods and improve the quality of the reconstructed speech.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2011年第12期2688-2692,共5页
Chinese Journal of Scientific Instrument
基金
国家自然科学基金(60971129)
江苏省普通高校研究生科研创新计划(CX10B_189Z
CX10B_191Z)
南京邮电大学青蓝计划(NY210031)资助项目
关键词
小波变换
压缩感知
矢量量化
线性规划
wavelet transform
compressed sensing
vector quantization
linear programming
