Imperial College London > Talks@ee.imperial > COMMSP Seminar > Multichannel source separation for robots audition

Multichannel source separation for robots audition

Add to your list(s) Download to your calendar using vCal

If you have a question about this talk, please contact Alastair Moore.

In this talk, I will deal with audio source separation problems using microphone arrays. The main application is robot audition within the Romeo project led by Aldebaran Robotics.

First, I will present a source separation algorithm based on a sparsity criterion using the parameterization of the quazi-norm lp, with p between 0 and 1. I will show how we can have promising separation results by increasing the sparsity constraint along the iterations.

Second, by further exploiting the multisensor aspect of our application (16 sensors are fixed on the head of a humanoid), I will present a source separation algorithm using a fixed (data independent) beamforming preprocessing. Beamforming reduces the reverberation and the noise effect and thus leads to better separation results. In order to take into account the influence of the head in the near sound manifold, we built the beamforming filters offline using the Head Related Transfer Functions (HRTF) of the robot. I will present and compare different versions of this two-step source separation algorithm in the iterative and the adaptive cases, and show how beamforming preprocessing improves significantly the separation results compared to the use of a blind source separation only.

Finally, I will give an overview of an Audio-Visual Database for Robot Audition and Source Separation, AV-DRASS, which I have recently recorded at Telecom ParisTech. This multi-sensor database was recorded using 2 microphone arrays, 2 cameras and one kinect for speakers tracking for different scenarios, up to 3 moving sources speaking at the same time. It will be available for download soon.

Speaker Biography

Mounira Maazaoui is a post-doctoral fellow within the Signal and Image Processing (TSI) department at Telecom ParisTech with the audio, acoustics and waves (AAO) group. She received the state engineering degree from the École Nationale d’Ingénieurs de Tunis in 2008 and the M.Sc. degree in Information Processing and Living Complexity (a co-dipolma with the M.Sc. Mathematics and Informatics of the university Paris Descartes) also in 2008. In 2012, she received the Ph.D. degree in the field of audio signal processing from Telecom ParisTech on the topic of Source Separation for Robot Audition within the Romeo project led by Aldebaran Robotics.

Her research interests include multichannel signal processing: source separation, beamforming, robot audition, and recently source separation aided by video as well as the development/implementation of real-time source separation algorithms.

This talk is part of the COMMSP Seminar series.

Tell a friend about this talk:

This talk is included in these lists:

Note that ex-directory lists are not shown.

 

Changes to Talks@imperial | Privacy and Publicity