31.aug, 14:00 ICT-315 "Large Vocabulary Continuous Speech Recognition for Turkish"
Tarkvarateaduse instituudi seminaril reedel, 31.08.2018 kell 14:00 esineb Ebru Arısoy Saraçlar (MEF Ülikool, Istanbul, Türgi) teemal "Large Vocabulary Continuous Speech Recognition for Turkish".
Seminar toimub ruumis ICT-315.
Turkish is an agglutinative language with rich morphology. This agglutinative nature makes Turkish a challenging language for Automatic Speech Recognition (ASR) applications. A common solution proposed for agglutinative languages as well as Turkish is to use sub-lexical units instead of words in ASR. In my talk, I will mainly focus on language modeling component of ASR system and explain our efforts in the last decade to improve Turkish Large Vocabulary Continuous Speech Recognition (LVCSR). Our efforts include using grammatical and statistical sub-lexical units in neural network language models in addition to conventional language models.