Human hearing inspires audio processing for better hearing aidsIn his doctoral dissertation, Tom Barker developed a sound source separation technique based on the way the human ear separates sounds. Better sound source separation enables hearing aids that function better in everyday, noisy surroundings.
By examining how humans process sounds, we can learn a lot about effective strategies for improving machine listening – computational approaches for making sense of sounds. One part of this field is sound source separation, or the separating of different sounds present within a recording. For his doctoral thesis, Tom Barker from the Laboratory of Signal Processing has been investigating various techniques for sound source separation.
One of the approaches Barker focussed on was a technique inspired by the processing that takes place inside the human ear. By representing the sound in this way, using a representation called the Modulation Spectrogram, sounds can be separated more effectively than in more traditional domains for sound processing in situations where no prior training data is available.
Tom Barker’s thesis also introduces a low-latency sound separation technique, particularly focussed for application in hearing aids. This technique was developed in partnership with the Eriksholm Research Centre in Denmark, a world leader in hearing research.
“Source separation is an important technology in this domain, since many hearing aids perform considerably worse in the presence of multiple sound sources. With more efficient source separation algorithms, hearing impaired listeners will have an easier time communicating in many common day-to-day situations,” Barker says.
“The technology would also be useful in smart-home assistants, musical and broadcast processing tools, and for speech enhancement in telecommunications.”
Public defence of a doctoral dissertation on Wednesday 24 May
MEng Tom Barker's doctoral dissertation in the field of signal processing, entitled “Non-negative factorisation techniques for sound source separation”, will be publicly examined at the Faculty of Computing and Electrical Engineering, Tampere University of Technology (TUT) in room TB109 in the Tietotalo building (address: Korkeakoulunkatu 1, Tampere, Finland) at 12 noon on Wednesday 24 May 2017. The opponents will be Professor Tom Bäckström (Aalto University, Finland) and Antoine Liutkus (Inria, Nancy, France). Professor Tuomas Virtanen from the Laboratory of Signal Processing at TUT will act as Chairman.
Tom Barker (30) comes from Norwich, England and works as a researcher within the Audio Group in the Laboratory of Signal Processing of TUT.
Further information: Tom Barker, tel. +358 40 218 1167, thomas.barker (at) tut.fi