Ning Bi
Yingyong Qi
Dept. of Speech and Hear. Sci., Bldg. 71, Univ. of Arizona, Tucson, AZ 85721
A speech enhancement system based on spectral substitution was developed for alaryngeal talkers. In this system, the spectra of alaryngeal speech were converted to those of normal speech on a frame-by-frame basis. Synthetic speech was generated with the converted spectral information. The codebook for the spectral conversion consisted of cepstral coefficients of speech utterances. The rules for the spectral conversion were developed using procedures of supervised and unsupervised codebook adaptations. The synthetic speech produced by this spectral substitution scheme was compared perceptually to the synthetic speech produced by a linear predictive analysis-synthesis system, where only the voice source was replaced. The synthetic speech was also compared to the original alaryngeal speech in terms of intelligibility, acceptability, and intonation. [Work supported by NIH.]