Third International Conference on Spoken Language Processing (ICSLP 94)

Yokohama, Japan
September 18-22, 1994

Comparative Study of Spectral Representations in Measuring the English /r/-/l/ Acoustic-Perceptual Dissimilarity

Kiyoaki Aikawa, Reiko A. Yamada

ATR Human Information Processing Research Laboratories, Kyoto, Japan

The talker dependency of the correct response rate (CRR) on English /r/-/l/ identification by Japanese listeners has already been reported. This paper shows that the talker dependency of the CRR can be explained by the acoustical dissimilarity (ADS) between an /r/ and an /I/ measured by the dynamic-cepstrum. The dynamic-cepstrum is a new spectral representation which simulates time-frequency forward masking. Nine spectral representations including weighted-cepstrum, mel-cepstrum, and delta-cepstrum were compared in terms of correlation between the CRR and the ADS. The ADS measured by the dynamic-cepstrum showed the best correlation with the CRR. The experimental results imply that Japanese listeners tend to identify /r/ or /I/ using the succeeding vowels affected by co-articulations.

Full Paper

Bibliographic reference.  Aikawa, Kiyoaki / Yamada, Reiko A. (1994): "Comparative study of spectral representations in measuring the English /r/-/l/ acoustic-perceptual dissimilarity", In ICSLP-1994, 2039-2042.