MODELLING THE PERCEPTION OF CONCURRENT VOWELS
Peter F. Assmann, Quentin A. Summerfield · 2024
The effects of noise on speech communication have been understood in broad terms for some time (10).However, only relatively recently have speech researchers begun to address the question of how listeners extract individual acoustic cues to phonetic categories from speech signals corrupted by noise (6).and how that process can be modelled to be independent of the spectra-temporal structure of the noise (17).A particular problem, both for human speech perception and for automatic speechrecognition, arises when the background noise is that of a competing voice.Three cues for segregating a single voice from a background of other voices have received attention: binaural information (9).fundamental frequency differences (3.15).and onset/offset esynchronies (6,16).In this paper we investigate a restricted case of the multi-sourcs recognition problem, where none of the above cues are available: that of two simultaneous, synthetic vowels presented monaurally with the same, fixed fundamental frequency, and their pitch pulses in synchrony .-Scheffers (15) reported that listeners could identify both members of such a pair of vowels with better than-chence accuracy.This rather surprising finding was confirmed in a later study by twicker (la) and was replicated in