Multilingual Data Selection for Low Resource Speech Recognition

Samuel Thomas, Kartik Audhkhasi, Jia Cui, Brian Kingsbury, Bhuvana Ramabhadran
2016 Interspeech 2016  
Feature representations extracted from deep neural networkbased multilingual frontends provide significant improvements to speech recognition systems in low resource settings. To effectively train these frontends, we introduce a data selection technique that discovers language groups from an available set of training languages. This data selection method reduces the required amount of training data and training time by approximately 40%, with minimal performance degradation. We present speech
more » ... cognition results on 7 very limited language pack (VLLP) languages from the second option period of the IARPA Babel program using multilingual features trained on up to 10 languages. The proposed multilingual features provide up to 15% relative improvement over baseline acoustic features on the VLLP languages.
doi:10.21437/interspeech.2016-598 dblp:conf/interspeech/ThomasACKR16 fatcat:rnumqswls5ek3mfbrscx5omjwy