This work is about the application of certain properties of the auditory system to computational speech processing. The goal is to reduce disturbing effects of background noise, with the underlying assumption that the biological model is better suited for the solution of these problems, compared to entirely 'technical' approaches. For noise detection and suppression, spectro-temporal patterns are generated from the waveform which reflect the representation of amplitude modulations in higher stages of the auditory system, and which allow for a distinction between speech and noise portions. For noise-robust feature extraction in automatic speech recognition systems, a psychoacoustical model of the auditory periphery is applied and investigated. Both algorithms are combined to further enhance the robustness in automatic speech recognition. <engl.>
Collected papers / 137. Meeting of the Acoustical Society of America and the 2. Convention of the European Acoustics Association: Forum Acusticum, integrating the 25. German Acoustics DAGA Conference Oldenburg, 1999 [4] Bl.
