Fourth European Conference on Speech Communication and Technology

Madrid, Spain
September 18-21, 1995

A Binaural Selectivity Model for Speech Recognition

Markus Bodden (1), Timothy R. Anderson (2)

(1) Ruhr-University Bochum, Lehrstuhl AEA, Bochum, Germany
(2) Armstrong Laboratory, Bioacoustics and Biocommunications Branch, Wright-Patterson AFB, OH, USA

Neural networks that employed unsupervised learning were used on the output of a binaural auditory model, know as the Cocktail-Party-Processor, to perform context-independent phoneme recognition. Experiments which compared the performance of the binaural model representation to that of a monaural version showed that the binaural model performed significantly better in terms of phoneme recognition accuracy under the conditions tested (low signal-to-noise ratios (SNR) and a small database of speakers). The binaural model representations' performance has approximately a 20 dB SNR advantage over the monaural representation.

Full Paper

Bibliographic reference.  Bodden, Markus / Anderson, Timothy R. (1995): "A binaural selectivity model for speech recognition", In EUROSPEECH-1995, 127-130.