next up previous contents
Next: WP1: Conclusion Up: Task 1.3: Baseline System Previous: Language modeling

Task 1.3: Future Developments

We are in the middle of the alignment/training process and the evaluation of the system is still producing a large word error. Some causes of this error are known. The speech database used is still rather small. The alignment/training process is not complete, too. Furthermore, the lexicon is not yet fully corrected. As these limitations progressively get eliminated, we expect to be able to approach the performances that are obtained for American English with recognizers of the same kind.

The second part of this work is based on the first 10 speakers of a setup set extracted from the new database. We started to build a new system based on a new lexicon and a new bigram language model. With the further development of this system and its extension to the complete database, we expect to obtain a good recognizer for the European Portuguese language.

The work done under this task will continue in the next year through the improvement of the training/alignment procedure and from the inputs from Tasks 2.2 and 3.3. Also the availability of a large speech and text database for Portuguese will allow us to develop a better system for Portuguese.


next up previous contents
Next: WP1: Conclusion Up: Task 1.3: Baseline System Previous: Language modeling
Christophe Ris
1998-11-10