In this article, a fast automatic segmentation and labeling system is presented. The system is capable of labeling speech originating from most languages without requiring extensive linguistic knowledge or large (manually segmented and labeled) databases of that language. The system comprises small neural networks for the segmentation and the broad phonetic classification. They were originally trained on one task, and automatically adapted to a new task. Due to the limited size of the neural networks, the segmentation and labeling strategy requires but a limited amount of computations, and the adaptation to a new task can be accomplished very quickly. The performance of our system on TIMIT and on the English, Danish and Italian portions of the EUROMO compares favourably to that of other systems reported in the literature.
Bibliographic reference. Vorstermanst, A. / Martens, Jean-Pierre / Coile, Bert Van (1995): "Fast automatic segmentation and labeling: results on TIMIT and EUROMO", In EUROSPEECH-1995, 1397-1400.