ETRW on Speech Processing in Adverse Conditions Cannes-Mandelieu, France November 10-13, 1992

Evaluation of Linear and Non-Linear Spectral Subtraction Methods for Enhancing Noisy Speech

A. Le Floc'h (1), R. Salami (2), B. Mouy (3), J-P. Adoul (2)

(1) Thomson-CSF, dispatched to the university of Sherbrooke
(2) Departement de Genie Electrique, Université de Sherbrooke, Sherbrooke, Québec, Canada
(3) THOMSON-CSF division RGS, Gennevilliers, France

We present in this paper a comparative study between several spectral subtraction methods for speech enhancement, with the aim of improving the performance of LPC-based low bit rate speech coders in noisy environments. We also describe an improved voice-activity detector which is able to better distinguish between unvoiced speech and noise at low SNRs. Incorporating the studied spectral subtraction methods in ACELP4800 and LPC10-E resulted in significantly improved performances. The studied algorithms are compared in terms of the LPC spectral distance between the uncorrupted and enhanced speech signals.

Full Paper

Bibliographic reference.  Floc'h, A. Le / Salami, R. / Mouy, B. / Adoul, J-P. (1992): "Evaluation of linear and non-linear spectral subtraction methods for enhancing noisy speech", In SPAC-1992, 131-134.