Fully-funded PhD Studentship in
Multi-microphone Distant Speech Recognition in Domestic Settings
The studentship will be supervised by Jon Barker, University of Sheffield, UK; and supported by Toshiba Research Europe.
This project will be developing new approaches to multi-channel distant microphone speech recognition for use in noisy domestic settings. The project will be part of a collaboration between Sheffield and Toshiba to develop systems for the forthcoming CHiME-5 and CHiME-6 challenges, i.e. speech recognition of real dinner party conversations taking place in real homes (http://spandh.dcs.shef.ac.uk/chime_challenge/
). Key challenges are the shortage of in-domain training data and the high levels of background noise (including overlapping speech sources). The project will initially investigate approaches for training data augmentation using models of room acoustics for noisy speech simulation. It will then target approaches for coupling multichannel signal enhancement and speech recognition using deep neural network architectures. The student will be working as part of a Sheffield-Toshiba team that has committed to the development of systems for entry into the CHiME speech recognition evaluations. It is expected that the student will make one or more extended visits to Toshiba's Cambridge Research Labs during the course of the project.
This PhD studentship is jointly supported by Toshiba Research Europe and the University of Sheffield. The studentship includes full fees and a bursary at the standard EPSRC-rate for 3.5 years. We would like the successful applicant to start no later than February 2018 but an earlier start date would be possible. Applications will be considered on a rolling basis until the position it filled.
*Eligibility*: The studentship is open to UK, EU and Overseas applicants.