Close
Speech Science Forum - Benedetta Cevoli & Bethan Thomas (Speechmatics)
23 March 2023, 4:00 pm–5:30 pm
Improved Language Coverage through Self-Supervised Learning in Automatic Speech Recognition
This event is free.
Event Information
Open to
- All
Cost
- Free
Organiser
-
Justin Lo
Abstract:
Speech interaction with digital technologies is becoming more a part of everyday life, and as we increase our use of voice commands, accessibility and fairness across the world’s languages becomes ever more important. At Speechmatics, we have introduced self-supervised learning in an effort to increase our global language reach. Some languages have tens of thousands of hours of labelled acoustic data, others have just a few hours. In this talk we will explore the benefits of sample efficiency that we see from self-supervised learning, and how it is enabling us to reach state of the art performance across both high- and low-resource languages.
About the Speaker
Benedetta Cevoli & Bethan Thomas
at Speechmatics
Benedetta is a Cognitive Scientist with expertise across the fields of social sciences and natural language processing (NLP) interested in making technologies work for everyone. She’s currently focussing on topics of AI Bias and inclusion in the settings of automatic speech Recognition as a Data Scientist at Speechmatics. She previously completed an interdisciplinary PhD between Psychology and Computer Science at Royal Holloway, University of London studying the interlinks between artificial language models and human language. She has published several peer-reviewed scientific articles across the field of psycholinguistics, artificial intelligence, and sleep and holds a BSc in Psychology from the University of Trieste, Italy.
Bethan is a Machine Learning Engineer specializing in speech technology. She has a background in linguistics, from where she moved into the field of speech and language processing. She has several years industry experience working on various projects including text-to-speech, automatic language evaluation, sign language generation and, most recently, automatic speech recognition. She is passionate about understanding and reducing bias in AI systems, as well as methods to utilize the huge power of self-supervised learning for speech.