ISCA Archive - Recent advances in ASR applied to an Arabic transcription system for Al-Jazeera
ISCA Archive Interspeech 2014
ISCA Archive Interspeech 2014

Recent advances in ASR applied to an Arabic transcription system for Al-Jazeera

Patrick Cardinal, Ahmed Ali, Najim Dehak, Yu Zhang, Tuka Al Hanai, Yifan Zhang, James R. Glass, Stephan Vogel

This paper describes a detailed comparison of several state-of-the-art speech recognition techniques applied to a limited Arabic broadcast news dataset. The different approaches were all trained on 50 hours of transcribed audio from the Al-Jazeera news channel. The best results were obtained using i-vector-based speaker adaptation in a training scenario using the Minimum Phone Error (MPE) criteria combined with sequential Deep Neural Network (DNN) training. We report results for two different types of test data: broadcast news reports, with a best word error rate (WER) of 17.86%, and a broadcast conversations with a best WER of 29.85%. The overall WER on this test set is 25.6%.