Second International Conference on Spoken Language Processing (ICSLP'92)

Banff, Alberta, Canada
October 13-16, 1992

A New Dual-Channel Speech Enhancement Technique with Application to CELP Coding in Noise

S. Nandkumar, John H. L. Hansen, Robert J. Stets

Department of Electrical Engineering, Duke University, Durham, NC, USA

In this study, low bit-rate speech coding in noisy acoustic environments is addressed. Speech coding based on various forms of linear predictive coding is known to be sensitive to additive background noise. This study investigates in detail a new auditory constrained enhancement (ACE) technique as a front-end enhancer for a 4800 bps CELP coder operating in noisy environments. A novel, dual-channel, auditory based constrained iterative enhancement scheme is developed and shown to improve quality over all classes of speech. Performance of the proposed enhancement-coding tandem is evaluated using objective speech quality measures for nine background noise conditions over a subset of the NIST/TIMIT database. Average objective quality measures indicate that CELP maintains the level of distortion for speech coded in white Gaussian noise, but further degrades speech coded in non-stationary colored noise. A significant quality improvement for the tandem is shown, which is consistent over all speech classes and most phonemes.

Full Paper

Bibliographic reference.  Nandkumar, S. / Hansen, John H. L. / Stets, Robert J. (1992): "A new dual-channel speech enhancement technique with application to CELP coding in noise", In ICSLP-1992, 527-530.