Analysis, synthesis, and perception of Japanese voiceless stop consonants based on a pole-zero model.

Abstract

The production process of voiceless stop consonants was represented by a non-stationary autoregressive moving-average (ARMA) process, and the trajectories of poles and zeros from the noise burst to the following vowel were extracted from Japanese utterances of CV syllables containing voiceless stop consonants /p/, /t/, and /k/, using the SEARMA method. In order to clarify the relationship between the extracted parameters and the cues for perception of voiceless stop consonants, perceptual tests were conducted using synthetic CV syllables generated under various control conditions for these parameters. The relationship between the perceptual cues of voiceless stops and the acoustical parameters (i.e., the spectra and the duration of the consonantal noise, the orders of the model, the vocalic transition) were revealed.