Projects

Tabula Rasa - EU FP7 ICT

Together with a team of 11 European and international partners, EURECOM researchers are working to highlight and quantify the threat of spoofing to state-of-the-art biometric systems and to pioneer new biometric spoofing countermeasures. The consortium is considering a wide array of different biometrics: physiological, behavioural, multimodal, even new and emerging biometrics such as those involving electro-physiological signals. We are considering both software-based and hardware-based approaches to liveness detection, challenge-response countermeasures and new recognition methods which are intrinsically robust against attack. Federico Alegre's work in speaker recognition is investigating the potential of speech quality assessment, higher-level supra-segmental features, spectral texture analysis and fused approaches to detect spoofed speech signals. We are also contributing to multimodal biometric countermeasures, where face and voice modes are combined. Such approaches offer strong potential for spoofing protection it is potentially more challenging to spoof two biometric modes simultaneously.

Our aim is that spoofing protection becomes inherent in the system design process. The ultimate goal is to see our research results integrated within new biometric products and technology. Together with our industrial partners the TABULA RASA project will improve biometric security and to build the next generation, spoofing-robust biometric systems.

Learn more about Tabula Rasa here.

Adaptable Ambient Living Assistant - European Ambient Assisted Living (AAL) Joint National Programme

The primary objective of the Adaptable Ambient LIving ASsistant (ALIAS) project is the product development of a mobile robot system that interacts with elderly users, monitors and provides cognitive assistance in daily life, and promotes social inclusion by creating connections to people and events in the wider world. ALIAS is embodied by a mobile robot platform for which Asmaa Fillatre is developing and integrating speaker diarization and speaker recognition functionalities.

Learn more about ALIAS here.

Speech quality enhancement for mobile terminals

Sponsored by Intel Mobile Communications, Christelle Yemdji is investigating new speech enhancement algorithms for mobile terminal applications. The managing of computational complexity is central to her work.  Speech enhancement algorithms such acoustic echo cancellation or noise reduction are conventionally optimised and operate independently of one another but function in tandem to deliver the same goal.  Christelle is investigating the combination of such approaches to improve speech enhancement performance while reducing computational complexity.  She is also developing novel, dual-microphone solutions.

Noise compensation for low signal-to-noise-ratios

Adrien Daniel is investigating signal processing solutions for speech enhancement in the context of mobile communications. His work is sponsored by Intel Mobile Communications. The project aims to improve mono-channel noise reduction algorithms for single speaker signals corrupted by non-stationary noise with a low signal-to-noise ratio. In such critical cases, even perfect knowledge of the noise power spectral density is not sufficient to ensure enhanced speech of acceptable quality when using conventional signal processing methods. Hence the work aims to develop novel perception-based optimizations which go beyond the limits of existing state-of-the-art solutions.

Multi-modal biometrics and co-training

Xuran Zhao is working on multi-modal biometrics and related machine learning techniques. His work is sponsored by the “Futur et Ruptures” research programme funded by Institut Telecom. Standard biometric systems use labeled data collected during enrollment to train client models, but the amount of training data is rarely sufficient to reliably represent the variation which occurs later during testing. Xuran is investigating semi-supervised learning techniques, especially co-training, to solve this problem. His is also working in developing new semi-supervised/unsupervised machine learning methods to deal with more general multi-view classification/clustering problems which are inherent to applications in multimedia data management.