Mon-3-1-2 Development of Multilingual ASR Using GlobalPhone for Less-Resourced Languages: The Case of Ethiopian Languages

Abstract: In this paper, we present the cross-lingual and multilingual experiments we have conducted using existing resources of other languages for the development of speech recognition system for less-resourced languages. In our experiments, we used the Globalphone corpus as source and considered four Ethiopian languages namely Amharic, Oromo, Tigrigna and Wolaytta as targets. We have developed multilingual (ML) Automatic Speech Recognition (ASR) systems and decoded speech of the four Ethiopian languages. A multilingual acoustic model (AM) trained with speech data of 22 Globalphone languages but the target languages, achieved a Word Error Rate (WER) of 15.79%. Moreover, including training speech of one closely related language (in terms of phonetic overlap) in ML AM training resulted in a relative WER reduction of 51.41%. Although adaptation of ML systems did not give significant WER reduction over the monolingual ones, it enables us to rapidly adapt existing ML ASR systems to new languages. In sum, our experiments demonstrated that ASR systems can be developed rapidly with a pronunciation dictionary (PD) of low out of vocabulary (OOV) rate and a strong language model (LM).