Third International Conference on Spoken Language Processing (ICSLP 94)
Distinguishing between the voiceless fricatives F and TH is a difficult problem for both natural speech and synthetic speech. We report the results of experiments and spectral analyses designed to find distinguishing acoustic characteristics of the voiceless fricatives F and TH. These experiments were also designed in consideration of our hybrid text-to-speech system, which combines formant synthesis with concatenated units from natural speech. In our system, the intelligibility of initial stops has improved dramatically in comparison with the formant-synthesizer-only version of our system, but F and TH are still highly confusable . In this study, we used only natural speech, and conducted perceptual experiments by using frication-only stimuli and stimuli combining frication with segments of the following vowel. The results showed that when a frication portion alone was presented, F was correctly identified more often than TH. When the frication portion with the entire following vowel was presented, the identification of F showed significant improvement, reaching more than 90% for the /a u/ vowel contexts, and for the /i/ context, increasing from 40 to 78%. By contrast, the identification of TH failed to show any significant improvement.
Bibliographic reference. Hata, Kazue / Moran, Heather / Pearson, Steve (1994): "Distinguishing the voiceless fricatives f and TH in English: a study of relevant acoustic properties", In ICSLP-1994, 327-330.