Effect of Pitch Averaging on the Quality of Natural Vowels

A pitch synchronous analysis was carried out over the vowel portions of the CVC utterances HAYED, HOD, HODE, and the stressed vowel portions of the sentence, “FEW THIEVES ARE NEVER SENT TO THE JUG,” recorded by a male speaker. The analysis provided formant frequencies, excitation parameters, and accurate pitch-period measurements. For each vowel, each measured period was replaced by a period obtained by averaging adjacent periods over a symmetric “window” producing a smoothed pitch contour. The speech samples were reconstituted with the vowels with smoothed pitch contours. Eight window sizes were selected for subjective evaluation, ranging from an imperceptible amount of smoothing to an amount sufficient to produce nearly monotone vowels. Subjective evaluation was carried out by means of paired-comparison tests between utterances with different window sizes, including those with natural-pitch contours. The results indicate that listeners are unable to discriminate between samples with natural-pitch contours and those with an amount of smoothing sufficient to eliminate most of the period-to-period fine structure in the contours. In a addition, it appears that modes of discrimination with respect to pitch smoothing are distinct for the two cases of isolated vowel sounds and sentence-imbedded vowel sounds.