The contribution of waveform interactions to the perception of concurrent vowels

Abstract

Models of the auditory and phonetic analysis of speech must account for the ability of listeners to extract information from speech when competing voices are present. When two synthetic vowels are presented simultaneously and monaurally, listeners can exploit cues provided by a difference in fundamental frequency (F0) between the vowels to help determine their phonemic identities. Three experiments examined the effects of stimulus duration on the perception of such "double vowels." Experiment 1 confirmed earlier findings that a difference in F0 provides a smaller advantage when the duration of the stimulus is brief (50 ms rather than 200 ms). With brief stimuli, there may be insufficient time for attentional mechanisms to switch from the "dominant" member of the pair to the "nondominant" vowel. Alternatively, brief segments may restrict the availability of cues that are distributed over the time course of a longer segment of a double vowel. In experiment 1, listeners did not perform better when the same 50-ms segment was presented four times in succession (with 100-ms silent intervals) rather than only once, suggesting that limits on attention switching do not underlie the duration effect. However, performance improved in some conditions when four successive 50-ms segments were extracted from the 200-ms double vowels and presented in sequence, again with 100-ms silent intervals. Similar improvements were observed in experiment 2 between performance with the first 50-ms segment and one or more of the other three segments when the segments were presented individually. Experiment 3 demonstrated that part of the improvement observed in experiments 1 and 2 could be attributed to waveform interactions that either reinforce or attenuate harmonics that lie near vowel formants. Such interactions were beneficial only when the difference in F0 was small (0.25-1 semitone). These results are compatible with the idea that listeners benefit from small differences in F0 by performing a sequence of analyses of different time segments of a double vowel to determine where the formants of the constituent vowels are best defined.