 |
Workshop on the Auditory Basis of Speech PerceptionKeele University, UK |
 |
|
Workshop on the Auditory Basis of Speech PerceptionKeele University, UK |
|
A series of studies concerning the human ability to discriminate vowels are reported. The first study established the thresholds for discrimination under optimal listening conditions for isolated vowels synthesized from a female talker. Extensions of that research have revealed several new findings. First, the stimulus factors fundamental frequency and consonantal context can significantly degrade formant frequency resolution. Second, changes in training protocols for listeners or the levels of stimulus uncertainty degrades formant resolution substantially. Another set of studies found that listeners with moderate, sloping hearing impairment had good resolution for Fl formants but poor resolution for F2 even when clearly audible. Several auditory models have been developed to account for significant variability due to formant and fundamental frequency observed for formant thresholds. Two major classes of models have been examined, both of which are based on spectral-difference metrics between the standard vowel and the vowel that is just discriminable at threshold. The first class used excitation patterns as the spectral representation and the second derives a specific loudness metric called A Sone. Both metrics were shown to account for the variability in formant thresholds, with the A Sone metric the most successful thus far.
Bibliographic reference. Kewley-Port, Diane (1996): "Psychophysical studies of vowel formants", In ABSP-1996, 148-153.