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

Banff, Alberta, Canada
October 13-16, 1992

Towards Handling the Acoustic Environment in Spoken Language Processing

Hynek Hermansky (1), Nelson Morgan (2)

(1) US WEST Advanced Technologies, Boulder, Colorado, USA
(2) International Computer Science Institute, Berkeley, California, USA

Most speech parameter estimation techniques are sensitive to linear distortions introduced by the frequency response of the communication environment. Such linear distortions appear as an additive constant in the log spectrum. RelAtive SpectrAI (RASTA) processing, in which time trajectories of temporal representation of speech are band-pass filtered, makes speech analysis less sensitive to steady-state spectral factors in speech. We describe this approach, compare it to a conventional environment-equalizing blind deconvolution technique, and give experimental results from pilot experiments.

Full Paper

Bibliographic reference.  Hermansky, Hynek / Morgan, Nelson (1992): "Towards handling the acoustic environment in spoken language processing", In ICSLP-1992, 85-88.