Open AccessACCELERATION IN STATE-OF-THE-ART ASR APPLIED TO A VIETNAMESE TRANSCRIPTION SYSTEM
Author(s) -
Nhut M. Pham,
Vũ Hải Quân
Publication year - 2019
Publication title -
journal of computer science and cybernetics (vietnam academy of science and technology)/journal of computer science and cybernetics
Language(s) - English
Resource type - Journals
eISSN - 2815-5939
pISSN - 1813-9663
DOI - 10.15625/1813-9663/34/4/13181
Subject(s) - vietnamese , computer science , word error rate , speech recognition , transcription (linguistics) , acceleration , hidden markov model , state (computer science) , natural language processing , artificial intelligence , linguistics , algorithm , physics , philosophy , classical mechanics
This paper presents the adoption of state-of-the-art ASR techniques into Vietnamese. To better assess these techniques, speech corpora in the research community are assembled, and expanded, making a unified evaluation material under the name VN-Corpus. On this corpus, three ASR systems are built using the conventional HMM-GMM recipe, SGMM, and DNN respectively. Experimental results crown DNN with the overall WER of 12.1%. In the best case, DNN even cut down to 9.7% error rate.