Speech and Audio Processing

EURECOM's Speech and Audio Processing Research Group was formed in 2007 and is today a growing team with interests in speaker diarization/indexing (segmentation and clustering of multiple/competing speakers), speaker recognition, multimodal biometrics and speech signal processing.

Current research topics include the investigation of new approaches to linguistic normalization to improve the performance of speaker diarization and short-duration speaker recognition systems.   Through a recent collaboration with the Technical University of Munich (TUM) we are investigating new solutions for the detection of overlapping speech, which is currently at the limits of the state of the art in several fields of speech and audio processing, and particularly in speaker diarization.  This work is partly funded through the European Adaptable Ambient Living Assistant (ALIAS)  project funded through the Ambient Assisted Living (AAL) Joint Programme.  In the EU FP7 ICT Tabula Rasa project, the group is developing new countermeasures to protect text-independent speaker recognition systems from spoofing. 

Through a growing number of sponsored research programmes funded by Intel Mobile Communications, we are investigating new approaches to non-linear acoustic echo cancellation, echo post-filtering, joint / synchronized echo control and noise compensation.  The first of these projects has just come to end and resulted in new approaches to non-linear loudspeaker modelling and pre-processing/compensation.

The group participated in the most recent internationally competitive Rich Transcription evaluations administered by NIST in the US and is co-organising the Tabula Rasa Spoofing Challenge to be held during the IAPR/IEEE International Biometrics Conference (ICB) in 2013, for which we are responsible for the challenge in speaker recognition.  Together with other colleagues at EURECOM, the group will host the EURASIP European Signal Processing Conference (EUSIPCO) to be held on the French Riviera in 2015.