The main purpose of the work here presented consists in applying a speech synthesis technique based on Hidden Markov Models (HMMs) to European Portuguese (EP), in order to build an HMMbased EP Text-to-Speech (TTS) System.
The HMM-based speech synthesis technique (HSS) was originally tested for Japanese, but it has already been tested for other languages, such as English and Brazilian Portuguese. In fact, HSS comprises a compact module, being able to be applied to different languages.
In the implemented TTS system, the speech waveform is generated by means of HMMs trained for an EP database, that are selected according to a set of automatically generated binary decision trees.
Cite as: Barros, M.J., Maia, R., Tokuda, K., Resende, F.G., Freitas, D. (2005) HMM-based european Portuguese TTS system. Proc. Interspeech 2005, 2581-2584, doi: 10.21437/Interspeech.2005-800
@inproceedings{barros05_interspeech,
author={Maria João Barros and Ranniery Maia and Keiichi Tokuda and Fernando Gil Resende and Diamantino Freitas},
title={{HMM-based european Portuguese TTS system}},
year=2005,
booktitle={Proc. Interspeech 2005},
pages={2581--2584},
doi={10.21437/Interspeech.2005-800}
}