Third Workshop on Spoken Language Technologies for Under-resourced Languages Cape Town, South Africa May 7-9, 2012

Syllable-Based and Hybrid Acoustic Models for Amharic Speech Recognition

Martha Yifiru Tachbelie, Solomon Teferra Abate, Laurent Besacier, Solange Rossato

Laboratoire d’Informatique de Grenoble, France

This paper presents the results of our experiments on the use of hybrid acoustic units in speech recognition and the use of syllable and hybrid acoustic models (AM) in morphemebased speech recognition. Although hybrid AMs did not bring improvement in speech recognition performance when words are used as dictionary entries and units in a language model (LM), we observed a significant word error rate (WER) reduction (compared to triphone-based systems) in morpheme-based speech recognition. Syllable AMs also led to a significant WER reduction over the triphone-based systems. It was possible to obtain a 3% absolute WER reduction as a result of using syllable acoustic units. Generally, our result shows that syllable and hybrid AMs are best fitted in morpheme-based speech recognition.

Index Terms: syllable-based acoustic models, hybrid acoustic models, morpheme-based speech recognition, Amharic

Full Paper

Bibliographic reference.  Tachbelie, Martha Yifiru / Abate, Solomon Teferra / Besacier, Laurent / Rossato, Solange (2012): "Syllable-based and hybrid acoustic models for Amharic speech recognition", In SLTU-2012, 5-10.