Intelligibility of speech signals spectrally compressed by a sampling-synthesizing technique

The purpose of this study was 1) to measure the intelligibility of speech signals processed with a frequency converter designed to effect a spectral compression by a sampling-synthesizing technique, and 2) to investigate the effect of a short training period on the perception of the speech signals so processed. The input signals were composed of 71 CVC words recorded by an American female speaker. The highest center frequency for the input of the 22 channels was 9500 Hz. The four compression ratios of the output to the fixed input frequency spectrum were 1.0, 0.7 0.5, and 0.4 for all 22 channels used. Eighty listeners were randomly selected and assigned to one of the four conditions of spectral compression. The intelligibility score of each condition of spectral compression was obtained from a multiple choice test of six possible choices for each stimulus. The average intelligibility score for all English phonemes was obtained for each condition of compression. The intelligibility of each vowel and consonant was also calculated for all conditions. Only 15 minutes of programmed training in listening to this processed speech was found to be significantly effective for improving perception in all four conditions.