Response time to a sentence verification task as a function of LPC narrow-band processing and bit error rate

Abstract

The comprehension of narrow-band digital speech with bit errors was tested using a sentence verification task. The difficulty of the verification task was varied by using predicates that were either strongly or weakly related to the subjects. (A toad has warts./A toad has eyes.) In addition, half of the sentences were about category relationships, and half were about property relationships. (A salmon is a bird./A camel has horns.) The test conditions included unprocessed speech and speech processed using a 2400 bit/s linear predictive coding (LPC) voice processing algorithm with random bit error rates of 0%, 2%, and 5%. In general, response accuracy decreased and reaction time increased with LPC processing and with increasing bit error rates. False sentences took longer to verify than did true sentences, but there were fewer errors for the false sentences. Weakly related true sentences and strongly related false sentences were more difficult than their counterparts. Interactions between sentence type and processing conditions will be discussed.