Every month, the joint laboratory invites external speakers to take part in seminars for its partners.
Clément Gaultier (Hearing Institute, FR): « Recovering speech perception in noisy environments for cochlear implant recipients: speaker extraction and multi-microphone approaches based on deep learning methods. »
Abstract:
Following speech in noisy and reverberant situations is difficult, in particular for cochlear implant (CI) recipients. Machine learning methods such as deep neural networks (DNN) have shown potential for enhancing speech in a variety of noisy environments. This talk will present data from two recent lines of work: A study on a speaker extraction method to steer speech enhancement strategies and an investigation into single- and multi-microphone algorithms on the joint task of denoising and dereverberation. The DNN algorithms were trained and tested on simulated sound scenes and their performance was assessed using objective measures and experimental listening studies with typically hearing and CI listeners. The results confirm significant and substantial improvements in speech intelligibility for these two approaches over both unprocessed noisy speech and baseline algorithms. These methods show promise to benefit users of CIs and hearing devices in challenging communication situations. Further steps and future challenges will be discussed.
Bio:
Clément Gaultier is a Postdoctoral Research Fellow at the Center for Research and Innovation in Human Audiology (CeRIAH), Institut de l’Audition, Institut Pasteur, Paris, holding a Pasteur-Roux-Cantarini fellowship. He obtained his PhD from the University of Rennes (France), carried out at Inria Rennes, on sparse models and algorithms for audio signal processing, and subsequently worked as a researcher at Orange Labs. He then transitioned to auditory sciences joining the Deep Hearing Lab at the MRC Cognition and Brain Sciences Unit, University of Cambridge (UK), where he studied deep learning-based speech enhancement for cochlear implant users. Clément is now fully established in hearing research and based at Institut de l’Audition, where his research investigates, with an interdisciplinary approach combining signal processing, machine learning, and auditory neuroscience, new methods to measure and restore auditory perception for individuals using cochlear implants and other hearing devices in realistic acoustic environments. He is also a member of the Executive Board of the Computational Audiology Network (CAN) and a visiting research scientist of the Deep Hearing Lab, an international group of researchers working to improve the perception of sound by people with hearing difficulties.
