University Homepage School Homepage School Contacts School Search

Informatics Inaugural Lectures

Speech Recognition and Understanding in Realistic Environments

Professor Steve Renals

Lecture Details:
5.15 pm, Thursday, 14 October 2004
Appleton Tower, Lecture Theatre 2
Crichton Street
Edinburgh EH8 9LE
The lecture will be followed by a reception, on open area of level 4 (NOTE CHANGE: reception has been moved from Appleton Tower concourse to Informatics level 4 foyer area).

Lecture Slides

Abstract

How can machines make sense of spoken language? People are able to process and recognise speech with remarkably few constraints on the speaker, the topic or the environment. Automatic speech recognition has made substantial progress for specific tasks, such as dictation in a quiet office and some telephone information services, but speech recognition in a natural environment, with multiple sound sources (for example, several talkers) is still an unsolved problem.

Speech technology is dominated by approaches based on hidden Markov modelling. Hidden Markov models have a simple hidden structure, enabling the efficient estimation of their parameters from huge amounts of training data. The relative success of such models for speech recognition has led researchers to apply them (or their close relatives) to tasks such as the extraction of information from, and the summarisation of, spoken language. However, despite their relative success, an important current research issue concerns the development of models with a richer hidden structure that may still be trained from large corpora. Furthermore, these models are well matched to multimodal data. The lecture will discuss these issues, with particular reference to work on making sense of multiparty meetings - a domain which brings these research challenges to the fore.

Biography

Steve Renals is Professor of Speech Technology in the School of Informatics at the University of Edinburgh, where he is also Director of the Centre for Speech Technology Research. His research is in the area of spoken language processing, based on modelling speech and language using well-founded mathematical and statistical models. His work has included many aspects of speech technology, in particular speech recognition, information access from spoken audio and models for multimodal data.

Before moving to Edinburgh in 2003, Steve Renals was Reader in Computer Science at the University of Sheffield. Previous to that, he held postdoctoral fellowships at the International Computer Science Institute, Berkeley and at the University of Cambridge. He holds a BSc from Sheffield, and an MSc and a PhD from Edinburgh.

Back to  Informatics Inaugural Lecture Series page.