Evidence and counter-evidence in speech perceptionResearch Student: Stuart Cunningham Supervisor: Martin Cooke
Speech communication rarely takes place in noiseless conditions. Additional sounds alter the spectral and temporal characteristics of the signal reaching our ears. Despite this, listeners manage to interpret the mixture of signals that confronts them in all but the most adverse conditions. Part of this competence may stem from the ability to organise this mixture of sounds according to their respective sources - a task variously referred to as auditory scene analysis or auditory grouping. Subsequent to any grouping process it is likely that listeners may have to solve a missing data problem, as for some spectro-temporal regions masking by competing sounds renders recovery of the complete speech signal problematic.
In experimental conditions listeners can 'restore' missing segments masked by sounds with sufficient energy. This criteria of sufficient energy suggests listeners may have access to mechanisms which can exploit counter-evidence (i.e. insufficient energy). Using perceptual experiments and computational models we hope to investigate the relative importance of evidence (present speech energy) and counter-evidence (masked speech energy).