Detection of frequency and intensity changes using synthetic vowels and other sounds

Abstract

Formant frequency transition detection thresholds In synthetic vowels were Investigated for their dependence on formant number, vowel type and duration of frequency transition. With transitions In final position It was found that F2 Is easier to detect than FI for all tested vowels when thresholds are expressed In terms of critical bands.
The proximity of neighbouring formants appeared to affect the thresholds; lower values were obtained for formants that were positioned In excess of a critical bandwidth from neighbouring formants. All thresholds exhibited a decreasing tendency with Increasing transition duration.
An excitation pattern model was used to compare thresholds from each of the experimental conditions. This was effective In normalising the data and confirming a masking hypothesis as an explanation for threshold differences.
Thresholds were also obtained for sinewave stimuli. These, as one would expect, proved to be superior to those obtained for formants (on average by 4.5:1} though the ratio was dependent on frequency.
Comparisons were made between frequency transitions for pure tones and DL data, a well-documented area of psychoacoustics. The data show a similar relationship, although discrimination thresholds were consistently lower than transition thresholds.
Frequency transition detection thresholds In Initial position were Investigated for their alleged Inferiority compared with final position transitions. This was confirmed universally by experiments on both formant and sinewave transitions. For the vowel stimuli Inconsistencies with a masking model were found. This suggested the possibility of a speech mode of perception that operated predominantly when speech cues were more powerful.
Finally, intensity transition detection thresholds (and difference limens) were obtained for synthetic vowel stimuli. Experimental conditions were analogous to those used for frequency transition detection. Similar functions were obtained, and vowel type proved to be unimportant. However, transitions of increasing intensity proved to be easier to detect than those of decreasing intensity.