Preference judgments comparing different synthetic voices
John S. Logan, David B. Pisoni · The Journal of the Acoustical Society of America · 1986
Speech from three text-to-speech systems was compared on the basis of listeners' preferences. Subjects heard a sentence produced by one system followed by the same sentence generated by another system. The task was to indicate which of the two voices they preferred and then furnish a confidence rating for the decision. Each subject was presented with 40 pairs of sentences. Ten subjects participated in each of the three pairwise comparison conditions: (1) DECtalk compared with MITalk, (2) DECtalk compared with the Prose-2000, and (3) MITalk compared with the Prose-2000. The overall relationship among preference, response time required for this decision, and the confidence rating was examined. Results indicated a direct relation between preference and intelligibility whereas the remaining measures did not appear to be systematically related to preference. These subjective judgments will be discussed in terms of the relationships among preference, naturalness, and intelligibility in the perceptual evaluation of natural and synthetic speech. [Work supported by AFOSR and NIH.]