Three methods of deriving an acoustic representation known as IMELDA have been tested for applications with degraded speech recorded in a helicopter. IMELDA computed for a specific speaker has been found most effective, producing high recognition accuracy for four speakers across all the experimental conditions. Other factors contributing to better recognition accuracy, such as noise thresholding and syntax architectures, are also described in the paper.
Cite as: Zwierzynski, D.A., Lefèbvre, C. (1992) Recognition of degraded speech with an IMELDA acoustic representation: a helicopter fly-by-voice project. Proc. ETRW on Speech Processing in Adverse Conditions, 191-194
@inproceedings{zwierzynski92_spac,
author={Dariusz A. Zwierzynski and Claude Lefèbvre},
title={{Recognition of degraded speech with an IMELDA acoustic representation: a helicopter fly-by-voice project}},
year=1992,
booktitle={Proc. ETRW on Speech Processing in Adverse Conditions},
pages={191--194}
}