Acoustic correlates of intonation in whispered speech

Abstract

Although it is generally accepted that fundamental frequency (F0) is the primary acoustic correlate of intonation, it is well known that intonational distinctions are easily perceived in whispered speech [W. Meyer-Eppler, J. Acoust. Soc. Am. 29, 104–106 (1957)]. Meyer-Eppler reports that analysis of whispered human speech shows different formant frequencies in question/statement oppositions of identical phoneme sequences. It seems plausible that these same formant shifts occur in voiced speech, perhaps as a result of larynx height change, or as a result of learned gestures to maintain the same phonetic quality as F0 varies. However, rule synthesis systems usually use F0 alone to signal intonation. In our experiments we have used speech synthesis by rule in which formant frequencies have been modified in a simple way dependent on the value of F0 as determined by the prosedic rules. When these formants are used with entirely voiceless excitation, subjects are able to make the same question/statement distinctions as with voiced synthesis, although the perceptual difference is less clear. These results may lead the way to achieving more natural-sounding intonation in voiced synthesis.