Sixth European Conference on Speech Communication and Technology

Budapest, Hungary
September 5-9, 1999

Multi-Lingual Speech Recognition Based on Demi-Syllable Subword Units

Tibor Fegyó, Péter Tatai

TSP Laboratory, Department of Telecommunications and Telematics, Technical University of Budapest, Hungary

Hungarian, unlike English, is an agglutinating language, so new, special methods are needed for speech recognition. The word dictionary could become very large due to its complex morphological system, so a suitable approach could be to use subword units as, eg., half (demi) syllables and language models. In this way most of the natural languages can be described, therefore this method can be applied in a multi lingual recognition system. We describe in the article a method to represent the demi-syllable segments efficiently. With this method the computation of the distortion measurement becomes fast. Our recognition results in speaker dependent case proved that the chosen method performs slightly better than DTW and requires sharply less computation time.

Full Paper (PDF)   Gnu-Zipped Postscript

Bibliographic reference.  Fegyó, Tibor / Tatai, Péter (1999): "Multi-lingual speech recognition based on demi-syllable subword units", In EUROSPEECH'99, 867-870.