The aim of the work described in this paper is to develop and evaluate the speaker normalization technique based on the test to reference speaker mapping. The method is suitable for uniform time-frequency representation of speech used in speech corrector systems.
The normalized spectrum is generated after the analysis by synthesis for the given utterance using the MBE (multiband excitation) coding. The MBE speech production model decomposes the short time spectrum into the spectral envelope and excitation spectrum. The model offers the convenient way for joint vocal tract and excitation characteristics mapping to the reference speaker and at the same time preserving the phonetically relevant information in the test speaker utterance.
Cite as: Ogner, M., Kacic, Z. (2000) Normalized time-frequency speech representation in articulation training systems. Proc. 6th International Conference on Spoken Language Processing (ICSLP 2000), vol. 3, 1113-1116, doi: 10.21437/ICSLP.2000-731
@inproceedings{ogner00_icslp,
author={Marcel Ogner and Zdravko Kacic},
title={{Normalized time-frequency speech representation in articulation training systems}},
year=2000,
booktitle={Proc. 6th International Conference on Spoken Language Processing (ICSLP 2000)},
pages={vol. 3, 1113-1116},
doi={10.21437/ICSLP.2000-731}
}