This paper presents the results obtained for the EVALITA 2011 Forced Alignment on Spontaneous Speech task, including some aspects explored for the generation of the lexicon. A classical system was used for the alignment and several tests were performed to determine the impact of frame shift size and the use of speaker adaptation on the accuracy of the alignment. Good segmentation results were obtained, the proposed system outperforming the other teams' systems. Furthermore, phonetic change rules were determined on the train set and employed in the alignment process, improving significantly the performance of the system.
展开▼
