Abstract: Sometimes a speech signal might be degraded by background noise, making it hard to understand the speech content. Speech enhancement and separation systems can be used to reduce the noise level and improve the quality and the intelligibility of a speech signal. The use of visual cues, such as mouth movements and facial expressions, might be beneficial for the systems and provide a substantial performance improvement.
In this session, participants will be introduced to recent advances in audio-visual speech enhancement and separation, which has a variety of different applications, including:
* hearing assistive devices;
* videoconference systems;
* noise reduction in live videos.
The objective of this session is to provide participants with the basic theoretical background that would allow them to understand and build state-of-the-art systems for audio-visual speech enhancement and separation.
MODULE 1 - Introduction to deep learning and audio-visual speech corpora.
Participants will familiarise themselves with the concept of deep learning with a specific focus on audio-visual speech enhancement and separation. In addition, some characteristics of common audio-visual speech corpora will be explained.
MODULE 2 - Audio-visual speech enhancement and separation systems.
Participants will learn the main components of state-of-the-art approaches for audio-visual speech enhancement and separation.
MODULE 3 - Examples and demos
Participants will have the possibility to learn about recent advances in audio-visual speech enhancement and separation systems and experience some demos of these systems.
Basic knowledge of signal processing and deep learning.
Bio: Daniel Michelsanti is an Industrial Postdoctoral Researcher at Demant and Aalborg University. He has a PhD in Electrical and Electronic Engineering obtained at Aalborg University. Currently, he is investigating cutting-edge technologies for next-generation hearing assistive devices, with the goal of improving the life quality of people with hearing loss.