This paper introduces recent activities at Speech Technology Laboratory on the study of the Lombard effect, ranging from acoustic, perceptual, and production analyses to speech recognition. To gain an understanding about the Lombard effect in perspective of improving performance of automatic speech recognizers, we 1) analyzed the acoustic-phonetic changes occurring in Lombard speech, 2) studied the influence of the Lombard effect on speech perception, 3) evaluated the influence of speech loudness on acoustic parameters, and 4) analyzed inter-articulatory relationships in vowel production in noisy and non-noisy conditions. Both acoustic and perceptual analyses suggest that the influence of the Lombard effect on male and female speakers is different. They also bring to light that, even if some tendencies across speakers can be observed consistently, the Lombard reflex is highly variable from speaker to speaker. The production study revealed ihat: 1) the type of noise influences speech production, and 2) the modification of speech production due to background noise is speaker-dependent and context-dependent. Based on the results of these studies, we also discuss some ways of dealing with Lombard speech variability in automatic speech recognition and propose the use of relational features, derived from our phonetic knowledge, to improve automatic speech recognition of Lombard speech.
Cite as: Junqua, J.-C. (1992) The variability of speech produced in noise. Proc. ETRW on Speech Processing in Adverse Conditions, 43-52
@inproceedings{junqua92_spac,
author={Jean-Claude Junqua},
title={{The variability of speech produced in noise}},
year=1992,
booktitle={Proc. ETRW on Speech Processing in Adverse Conditions},
pages={43--52}
}