Deep Learning for speech enhancement using in-ear transducers

Au LMSSC, Paris, le 6 janvier 2022 à 11h

Julien Hauret
Doctorant, LMSSC, Cnam, Paris

After a brief presentation of my academic background, I will expose the ins and outs of my PhD. It mainly consists in improving the intelligibility of speech captured by an in-ear microphone developed by ISL. This unconventional captation device, combined with an active hearing protection, allows to capture the vocal signals emitted by a speaker while eliminating all external noise pollution.

However, the acoustic path between the mouth and the transducers is responsible for a total loss of information above 2 kHz. We are therefore dealing with a bandwidth extension problem. Deep learning methods will be used for the reconstruction of high frequencies instead of the source-filter model which is not able to restore missing information.