Speech intelligibility modelling in noisy rooms

Speaker: Dr Mathieu Lavandier, University of Lyon, France
Date: Monday 1 August 2016
Time: 2:00pm – 3:00pm
Location: Australian Hearing Hub, Level 4,  National Acoustic Laboratories, Denis Byrne Room


Speech intelligibility in noise is improved by having two normally-functioning ears. These two ears allow for spatial release from masking: a noise source is less masking for speech if the target speaker and the noise source are at different positions. The underlying unmasking mechanisms are unfortunately impaired by reverberation, the sound reflections in the room surrounding the sources and listener. If the “noise” source is a competing voice, the listener can rely on other acoustical cues to segregate it from the target voice. Modulations in the temporal envelope of the competing voice allow for a better understanding of the target, the listener benefiting from the temporal “gaps” in the masker to access the target at a better signal-to-noise ratio. Again, reverberation impairs this mechanism by partly filling in the masker gaps. A difference in fundamental frequency between the competing voices can also be used by the listener to improve segregation. These different perceptual mechanisms involved while listening to speech in noisy rooms will be describe through the presentation of an intelligibility prediction model, which will be systematically confronted with behavioural data. This type of model is an interesting tool to investigate speech understanding in cocktail-party situation. One of our aim is to further develop this tool to take into account the influence of hearing impairment and hearing aids in realistic situations.


Mathieu Lavandier is a senior researcher at the University of Lyon in France and is a visiting associate in the Department of Linguistics at Macquarie University until mid-January. He is based at NAL. His research concerns sound perception in rooms, involving psychoacoustics, room acoustics and speech intelligibility modelling. His visit is partly funded by a visiting research fellowship from the Faculty of Human Sciences.