Quality Judgments of Spliced Synthetic Telephone Numbers
Joseph P. Olive, Lloyd H. Nakatani · The Journal of the Acoustical Society of America · 1971
Speech synthesis parameters (f0, f1, f2, f3, amplitude) of digits spoken in isolation and of telephone numbers were extracted. The parameters of the isolated digit were spliced together and resynthesized to generate the same telephone numbers as those analyzed. The parameters of the spliced numbers were manipulated in the following manner. (1) Timing: the duration of the spliced digits were equated to the duration of the corresponding digit in the natural number by a linear compression; (2) Pitch: the over-all pitch contour of the natural number was copied; (3) Formants: the formants of the spliced digits number were replaced by the formants of the natural number. These manipulations were done in all possible combinations, and the different versions were judged in three different tests. The listeners ordered the different versions of each number according to preference and comprehensibility, and rated their naturalness. A multidimensional scaling analysis revealed that (1) pitch was the dominant variable, and timing was negligible; (2) preference and comprehensibility were not equivalent; (3) naturalness depended primarily on pitch; (4) copied formants can detract from the quality of the speech; and (5) although results differed for different numbers, the trend was similar.