Department of Communication and Neuroscience, University of Keele, GB
A group at Keele University has been active in speech research for some thirty years. Work is currently in progress in the analysis and perception of speech and in models of the auditory system, the latter arising through co-operation with Professor E.F Evans' auditory physiology group. Experience includes participation in Esprit Project 2101 'ARS: Adverse-environment recognition of speech' led by CSELT, Turin, EU Science Projects SC1.0044.C(H) 'Investigation of speech recognition based on a physiological model of the auditory system' with ICP Grenoble and SC1-CT92-0786 'Modelling and perception of speech gestures' with ENST, Paris and the Free University of Brussels, and HCM Project SPHERE (CHRX-CT93-0098) led by Sheffield University.
The role of Keele University is to use their experience in speech perception, speech analysis and auditory modelling in SPHEAR themes 1, 2 and 3. The perceptual studies in theme 1will be used to determine the limits of perceptual time shifts acceptable to human listeners and these will be used to guide the recognition studies carried out at IDIAP. The background to this is that recently Greenberg [Greenberg 95] has shown that speech is understandable when different spectral channels are shifted up to 100ms in time with respect to each other. At Keele, Ainsworth [Ainsworth 1995, see below] found that in vowel-vowel transitions the temporal difference limen was of the order of 50ms. On the other hand, temporal coherence has been shown to play a role for event detection and Grenoble shown its role for the localisation of phonetic information [Morris et. al. 93]. The Amplitude Modulation maps developed at Keele and Grenoble will be used in theme 2 to explore 'attention' in ASA in conjunction with Sheffield. The results of the noise band masking studies in theme 3 will be compared with the experiments on the recognition of degraded speech at Sheffield and cocktail party recognition at Bochum. The finding behind the noise-masking work is that continuous speech has less effect on speech intelligibility than noise coincident with speech sounds, but a short burst of noise preceding a syllable has an even more deleterious effect [Ainsworth, 95b]. This phenomenon is to be further explored using noise in different spectral bands in task 3.6. Preliminary studies have shown the effects of broadband noise on plosives and vowels. The effects of narrow band noise and tones on these and also on approximants will be studied.
There will be co-operation with Grenoble, Sheffield, IDIAP and Bochum as outlined above.