Information transmission analysis of the effects of synthetic stop consonant duration modification on speech perception

摘要

Speech perception studies for hearing-aid users continue to mostly use speech recognition score for analysis of the listening test results. Although useful, this measure does not consider the relatedness of stimuli and responses and it could be affected by a subject's bias. On the other hand, information transmission analysis provides a measure of covariance between stimuli and responses and reduces the effect of subject bias for large sample sizes. The stimulus-response confusion matrix obtained in each experimental run can be used to obtain the relative information transmitted. The transmission performance in the context of specific features can be obtained by applying the relative information transmission measure to the matrices derived from the original confusion matrix by grouping the stimuli in accordance with those features. The effect of consonant duration modification on the recognition of stop consonants in the consonant-vowel context is presented. Synthetic speech stimuli were generated in which the formant transition duration, voice onset time, and burst duration were independently altered to study their individual effects on consonant recognition. The effect of altering these acoustic segments on the transmission of place and voicing features are investigated using information transmission analysis. The results suggest that at higher noise levels simulating increased hearing impairment, a formant transition duration modification of up to about 50% may be combined with burst duration modification to yield better performance.