Vowel identification based on amplitude modulation.

Abstract

This study investigated the extent to which flat-spectrum harmonic complexes could be identified as one of six vowels when three pairs of successive harmonics, located at the first, second, and third formant frequency values, were amplitude modulated. In experiment 1, the amplitude modulation (AM) rate was at or close to 10 Hz. In condition 1, all components were added in cosine phase, and the 10-Hz AM was in phase for all "formants." Performance improved monotonically with increasing modulation index, m. In condition 2, m was fixed at 0.5 and the level of each background harmonic was varied randomly (roved) from stimulus to stimulus. Even a rove range of only +/- 2 dB reduced scores considerably. Condition 3 was like condition 1, but with components added in random phase. Performance was very poor for all modulation indices. This suggests that subjects were unable to use momentary differences in level between formant and background harmonics, and supports the idea that, for cosine-phase stimuli, they were using information from the low-amplitude portions ("valleys") of the cochlea-filtered waveforms. In further conditions, the components were added in cosine phase and the AM had a different phase and/or different rate (10, 16, and 24 Hz) on the different formants. Scores were very similar to those obtained when the AM was identical for all formants. In experiment 2, the AM rate was at or close to 2 Hz. When all formants were modulated in phase at 2 Hz, very good performance was found for components added in cosine phase, and performance was essentially unaffected by making the AM different in rate and/or phase across formants. When the components were added in random phase, performance was well above chance when the formants were modulated in-phase at 2-Hz, but worsened markedly when the modulation differed in rate and/or phase across formants. Randomizing the level of each background harmonic caused performance to deteriorate and to become similar for cosine-phase and random-phase stimuli. Performance deteriorated further when the AM differed in phase across formants. The results suggest that, for a 2-Hz modulation rate, and when information from the valleys is not available, performance depends on momentary increases in level of the formant harmonics relative to the background.