Perceptual and articulatory noise in vowel imitation
Gautam K. Vallabha, Betty Tuller · The Journal of the Acoustical Society of America · 2001
Speakers exhibit systematic errors when asked to imitate self-produced vowels [B. Tuller and G. Vallabha, J. Acoust. Soc. Am. 107, 2855 (2000)]. These results were replicated (3 speakers, 45 systematically distributed self-produced vowel targets, 10 imitations per target) and compared to a simulation of the effects of articulatory and perceptual noise. Articulatory noise was modeled using the Haskins Articulatory Synthesizer. The tongue body center (TBC) was interpolated between the configurations for /i/, /a/, and /u/ and 90 of the intermediate configurations were perturbed by adding independent Gaussian noise (sd=1 mm) to the x and y coordinates of the TBC. Perceptual noise was modeled by (1) transforming each of the 90 intermediate vowels to F1/F2 bark (‘‘perceptual’’) space and (2) adding independent Gaussian noise (sd=vowel discrimination limen=0.28 barks) to each bark coordinate. The models did not match the directions of the imitation errors made by the speakers. However, the distribution of the error magnitudes, relative to the mean imitation, was approximately Rayleigh (0.28) across both the models and the speakers. These results indicate that while speech perception and/or production may have intrinsic Gaussian noise, this noise is not responsible for the systematicity of the imitation errors. [Work supported by NIMH.]