next up previous contents
Next: WP3: CONCLUSION Up: Task 3.3: Language Model Previous: Decomposition Results

Task 3.3: Future Development

Work is in progress to evaluate the LSA and mixture model approaches on the Broadcast News LVCSR task. We hope to report results by the review meeting. We plan to evaluate the Portuguese language models (both word- and morpheme-based) using the continuous speech recognition system we are developing for the Portuguese language. In order to provide a fair comparison with the baseline system, the components must be recomposed into their compound words. We are trying to found a suitable method for that, based on the modification of some probabilities in the language model.

Christophe Ris