Ikerbasque Researcher: Martin Cooke
The study of human speech communication
You work on computational auditory scene analysis, active hearing, second language acquisition, robust automatic speech recognition and speech perception. What is your main research focus now?
While these areas are often treated as distinct topics and researched by different scientific communities, what unites them for me is the study of human speech communication, which lies at the intersection of psychology, linguistics and engineering. We lack even very basic knowledge about how listeners are capable of conversing in such a wide range of situations. Most of what we do know comes from laboratory studies with clean speech, ideal listeners and artificial situations, but the real world contains lots of noise and reverberation, listeners who may be communicating in a second or third language and forms of speech which change depending on the context. My main interest is in how listeners cope with these less-than-ideal but actually very normal situations. One of my current projects involves a detailed examination of consistent confusions i.e. stimuli which a majority of listeners hear in the same but incorrect way. The goal is to construct a “chemistry” of speech perception, which explains why speech and noise react in certain ways. To achieve this, my target is to elicit responses from more than a hundred thousand listeners using what is probably the largest online speech perception experiment (http://www.thebiglisten.org.uk ), currently for English speech but soon for Spanish and Basque too.
You arranged the first speech separation challenge with researcher Te-Won Lee (UCSD, USA). What is it all about?
The challenge brought together scientists from many countries to tackle the problem of recognizing speech from two talkers speaking simultaneously. This is something that human listeners are quite capable of doing but is a tough task for computers. The purpose of such a challenge was not just to find out what techniques work best, but to inspire new collaborative research and to get talented young researchers interested. The challenge was won by a team from IBM USA.
You are now working as a Ikerbasque Research Professor at the University of the Basque Country. What has been the main change between the work at the University of Sheffield and the work you are doing now? Why did you decide to come to the Basque Country?
I continue working towards the goal of trying to understand the fundamentals of speech perception in noise. It is my impression that noise levels tend to be higher here than in the UK, so the challenge for listeners is more severe! More seriously, working in the Basque Country means that I have access to listeners with a richer linguistic experience than in the strictly monolingual UK, which benefits my research into computational models of second language acquisition. Coming to the Basque Country was an easy decision for me, since my immediate family were already here and for the first time we were able to live in the same country. I was also fortunate to be able to convince several of my doctoral students to move with me from Sheffield and I’d like to attract other talented researchers to work here, taking advantage of what are very low fees compared to countries like the UK. Above all, it is clear from other recent developments such as the new centre for language and cognition in San Sebastian that the Basque Country is set to become a very important nucleus in speech, hearing and linguistics in the coming years.
Share this:
