Discrimination of Vowels with a Multi-finger Tactual Display

The present study compared the performance of an ideal observer and a human participant in a vowel discrimination task using a speech-to-touch coding scheme designed for a three finger tactual display. The coding scheme extracted speech features and presented them as high-frequency vibrational and low-frequency motional waveforms. The high-frequency vibrations presented crude spectral information from three distinct speech bands on three fingerpads of the left hand. The same information was presented, redundantly, by the low-frequency waveforms. Performance of the ideal observer, where only high-frequency vibrational signals were considered, was evaluated by a signal detection theory using several tokens of a pair of vowels. Results showed that the acoustic cues corresponding to the first two formants were sufficient for discrimination of a seven vowel stimulus set. The participant was then tested in an absolute identification task with 640 tokens of ten non-diphthong vowels spoken by two female speakers. Both high- and low-frequency waveforms were presented to the participant. Discrimination scores for each pair of vowel were similar to the best scores obtained with the ideal observer indicating that the coding scheme was effective and the participant acted like an ideal observer.