Some effects of intonation contour on sentence intelligibility

Abstract

This experiment was designed to measure the effects of pitch movement on sentence intelligibility. A source-filter synthesizer was used to generate three synthetic versions of 60 sentences drawn from the TIMIT multi-talker speech database: (1) an original pitch (OP) condition in which the fundamental frequency (F0) contour matched that of the original utterance, (2) a monotone pitch (MP) condition in which F0 was held constant at the median value measured from the original utterance, and (3) an inverted pitch (IP) condition in which the F0 contour was reflected around the median F0 value (i.e., pitch rises were changed to pitch drops, and vice versa). Results from 30 listeners showed a small but statistically reliable drop in intelligibility from the OP condition to either the MP or IP condition, with no difference between the MP and IP conditions. A second group of 22 listeners was tested on the same task, but with overall sentence intelligibility reduced by running all signals through a 2-kHz low-pass filter. As with the unfiltered signals, intelligibility was reduced for the MP and IP conditions relative to OP; however, the decrements in intelligibility were somewhat larger for the filtered signals, and inverting pitch caused a larger intelligibility decrement than flattening pitch.