Login
Talks & Seminars
Title: Towards Automatic Speech Recognition for Low Resource Languages
Dr. Preethi Jyothi, University of Illinois at Urbana-Champaign
Date & Time: January 11, 2016 11:30
Venue: Conference Room, C Block, 01st Floor, Department of Computer Science and Engineering, Kanwal Rekhi (KReSIT) Building
Abstract:
The last decade has witnessed tremendous technological advances that have brought machine learning ideas to major real world applications. Automatic speech recognition (ASR) is one such application that has been a long-standing challenge. ASR for low resource languages is of particular significance as it has the potential for revolutionizing access to technology, especially in developing economies. However, mainstream approaches for ASR do not adequately address various challenges arising from lack of data and other resources. In this talk, I will describe several directions in my research that are motivated by these questions. One major challenge in developing ASR technologies for low resource languages is the difficulty in acquiring transcriptions for speech. As a potential solution for this problem, we introduced the idea of acquiring candidate transcriptions for speech in low resource languages from native speakers of a different (possibly related) major language. We have obtained promising results that demonstrate the potential of using transcriptions from such “mismatched transcribers.” I will also describe new statistical pronunciation models based on articulatory features (movement of lips, tongue, vocal cords etc.), motivated by their potential for generalization across dialects and languages. I will also briefly discuss new language modeling techniques motivated by the challenges in low-resource ASR.
Speaker Profile:
Dr. Preethi Jyothi is a Beckman Postdoctoral Fellow at the University of Illinois at Urbana-Champaign. She obtained her Ph.D. in computer science from The Ohio State University, in 2013. She obtained a B.Tech from the National Institute of Technology, Calicut, India, in 2006 where she was awarded the Institute Gold Medal in computer science. Her work on statistical learning methods for pronunciation models received a Best Student Paper Award at Interspeech, 2012. Based on her work on discriminatively trained language models, she was invited to spend the summer of 2012 at Google Research, Mountain View to work on large-scale language models. More recently, she co-organized a research project at the prestigious 2015 Jelinek Summer Workshop on Speech and Language Technology.
List of Talks

Webmail

Username:
Password:
Faculty CSE IT
Forgot Password
    [+] Sitemap     Feedback