The speech enhancement method proposed in this article is able to separate concurrent speech signals. This Cocktail-Party-Processor simulates - as far as possible - the binaural processing of the human auditory system. It therefore shows no principal restrictions with regard to sound-field or signal characteristics. The algorithm is designed to separate signals from spatially distributed sound sources. A binaural model is used to analyse the spatial distribution of the sound field. It produces neural excitation patterns that enable the system to estimate positions of sound sources and their energies. These estimates can be used to control a filterbank to extract the desired sound source. In order to evaluate the performance of the system the comprehensibility of interfered and processed speech have been measured. Results of tests with hearing impaired subjects are presented. Finally, the profit of using this system as a preprocessing unit for speech recognition technology is discussed.
Cite as: Bodden, M., Blauert, J. (1992) Separation of concurrent speech signals: a cocktail-party-processor for speech enhancement. Proc. ETRW on Speech Processing in Adverse Conditions, 147-150
@inproceedings{bodden92_spac,
author={Markus Bodden and Jens Blauert},
title={{Separation of concurrent speech signals: a cocktail-party-processor for speech enhancement}},
year=1992,
booktitle={Proc. ETRW on Speech Processing in Adverse Conditions},
pages={147--150}
}