New Master internship position with ADASP: Deep learning for joint detection and tracking of mobile sound sources via a moving microphone array
By M. Fontaine, S. Essid, G. Richard
Our group is hiring a Master intern on the topic “Deep learning for joint detection and tracking of mobile sound sources via a moving microphone array”.
- Date: March 2022
- Supervisors: Mathieu Fontaine, Slim Essid, Gaël Richard
- Contacts: email@example.com, firstname.lastname@example.org, email@example.com
- Salary: depending on the CV
Problem statement and context
Detection, localization and tracking of mobile sources in a noisy environment (urban noise, reverberation …) remains a complex research topic, hardly addressed in the machine listening community, which can lead to many applications (vehicle tracking, audio monitoring …). The challenge becomes even trickier when the microphone array is embedded in a mobile system such as a moving vehicle. Multiple interferences caused by the movement of the microphone array will indeed disrupt the data acquisition and defeat the classical tracking approaches.
In order to address such a complex problem, the topic will be oriented on hybrid deep learning approaches informed by audio signal models. We will focus on the problem of automatic identification of sound sources together with their tracking when the microphone array is in motion, by combining state-of-the-art deep networks (see DCASE) with tracking techniques. The intern will first have to generate an appropriate database for deep learning, partly using audio synthesis techniques. She/he will then work on developing an efficient deep neural network architecture in order to perform both localization and tracking of sound sources with the aforementioned constraints.
Required profile and additional comment
- The intern should have a good knowledge in deep neural networks, signal processing theory and Python.
- This internship proposal is linked to a PhD thesis project.