TECHCOMB

Volume 2 : Issue 2

Sparse Representation and Compressed Sensing of Speech via Segment MP

Authors : Li Zeng, Liang Chen, Xiongwei Zhang, Jibin Yang, Jianjun Huang

image

Abstract:

This article presents a Compressed Sensing (CS) based speech processing framework. Our main contributions are twofold: 1) a parametric incoherent dictionary is derived via Karhunen-Loeve Expansion to sparsely represent the signal. The parameter of the dictionary is adaptively estimated with each frame of speech; and 2) a new measuring and reconstruction pattern of CS, referred to as Segment MP (SegMP), is devised to implement the compressed speech signal sensing. We show that this dictionary is well adapted to the representation of speech signal since it enables smaller sparsity index. Extensive experiments demonstrate that the presented framework renders higher reconstruction accuracy while requires lower communication overhead in comparison to state-of-the-art CS schemes. It exhibits notable robustness and suggests a promising potential of further research on compressed speech signal sensing.