Distributed Compression of Correlated Audio Sources

Abhik Majumdar
(Professor Kannan Ramchandran)
Intel Corporation

We address the problem of compressing audio sources that are noisy filtered versions of the same audio signal. Exploiting the correlations between the remote sources will lead to better quality while maintaining the same transmission rate, as compared to a scheme that neglects these correlations. The objective is to develop algorithms for distributed compression of these correlated audio sources to attempt to achieve the gains predicted in theory [1]. The algorithms are based on the distributed source coding using syndromes (DISCUS) [2] framework and incorporate the use of perceptual masks, which are common in the compression of audio signals. We are currently in the process of implementing our algorithm. Initial results have been very encouraging.

We expect our algorithms to operate in a very power- and bandwidth-constrained environment. As such, our algorithms are designed to operate in medium to low bit rate regimes. Furthermore, we allow little or no communication among the sensors. Since the sensors may also have stringent power requirements and low computing power, the algorithms need to have sufficiently low complexity and also need to be optimized for the specific architectures of the audio sensors.


Figure 1: System set-up

[1]
A. D. Wyner and J. Ziv, "The Rate Distortion Function for Source Coding with Side Information at the Decoder," IEEE Trans. Information Theory, Vol. 22, January 1976.
[2]
S. S. Pradhan and K. Ramchandran, "Distributed Source Coding Using Syndromes (DISCUS): Design and Construction," Proc. IEEE Data Compression Conf. Snowbird, UT, March 1999.

More information (http://www.cs.berkeley.edu/~abhik) or

Send mail to the author : (abhik@eecs.berkeley.edu)


Edit this abstract