Consonant confusions by human listeners for loud speech.

Abstract

Loud speech stimuli were collected from three males and three females saying the 16 CV syllables of Miller and Nicely (1955). Headphone sidetone was mixed with 80 dBA pink noise. Talkers were instructed to speak loud enough to hear their voices over the masker. This resulted in average speech levels of 100 dBA at 1 ft. Only the loud speech was recorded. The intratalker speech levels were controlled within 2 dB. Seven subjects listened to the loud speech at a ‘‘comfortable’’ level, in quiet and mixed with multitalker babble. Subjects averaged 82% correct in quiet and 53% in babble. There were no significant performance differences for male and female talkers. Fricatives accounted for most errors with errors across all phonetic dimensions in both quiet and babble. In babble, subjects correctly identified voiceless stops 61% of the time but correctly classified stimuli as voiceless stops (place errors only) 92% of the time. Also in babble, nasals and voiced stops were confused 15% of the time; these errors never occurred in quiet.