Synthetic manipulation of prosodic cues: Effects on intelligibility and lexical segmentation in connected speech

The present experiment tested the hypothesis of a trading relation among the prosodic cues which signal lexical segmentation in connected speech. Seventy-five listeners were presented with 40 six-syllable utterances for a transcription task. The stimulus phrases were resynthesized to form 5 conditions: (1) resynthesized normal speech; (2) flattened F0 contour; (3) equalized vowel durations; (4) reduced vowel tokens; and (5) combined F0 flattened, vowel duration, and vowel reduction. All manipulations resulted in reduced intelligibility and errors in lexical parsing, with the resynthesized and combined conditions defining the performance ceiling and floor, respectively. The patterns of lexical boundary errors were analyzed in relation to the predictions of current lexical segmentation hypotheses. Within the individually manipulated conditions, the loss of vowel quality information was the most detrimental to signal intelligibility, and the equal vowel durations resulted in the greatest overall number of lexical boundary errors. The patterns of lexical boundary errors largely conformed to predictions, except for the vowel identity condition. The data suggest the existence of trading relations among the prosodic cues that listeners use to segment connected speech.