Dalle Molle Institute for Perceptual Artifiacal Intelligence, 
Martigny, CH

Experience and competence

The Dalle Molle Institute for Perceptive Artificial Intelligence (IDIAP, http://www.idiap.ch), located in Martigny (Valais, Switzerland), is a non-profit research institute affiliated with the Swiss Federal  Institute of Technology in Lausanne (EPFL) and the University of Geneva. The IDIAP research activities are primarily concentrating on automatic speech and speaker recognition, computer vision and machine learning. IDIAP's contributions to the current project are in state-of-the-art speech recognition systems (both HMM and HMM/ANN-based systems) and in multi-band/multi-stream speech recognition.

Role in the project

Prof. Hervé Bourlard will act as local coordinator.
IDIAP’s main technical contributions are in Theme 3 (speech recognition using auditory scene and speech production models) and in Theme 4 (speech recognition applications). IDIAP is supplying several speech training and recognition software, including new multi-stream software, new HMM2 training/recognition software, and Dynamic Bayesian Network software.

Research linkages

Thanks to SPHEAR, IDIAP has established good relationships with most of the HOARSE partners, and was particularly active (and interested) in exchanging young researchers.

Principal research personnel, expertise and involvement

  • Prof. Hervé Bourlard, Director of IDIAP, Leader of the IDIAP Speech Group, and Professor at the Swiss Federal Institute of Technology at Lausanne. Involvement 6 man-months.
  • Dr. Andrew Morris, Ph.D. (Grenoble), expertise in missing data and multi-stream techniques, Involvement 12 man-months.
  • Dr. Samy Bengio, Ph.D. (University of Montreal), Leader of the IDIAP Machine Learning group, expertise in machine learning, including HMM. Involvement 3 man-months.


    Recent publications

  • Morris, A., Hagen, A., Glotin, H., and Bourlard, H., “Multi-stream adaptive evidence combination for noise robust ASR,”  Speech Communication (Elsevier, North Holland), Vol. 34, Nos.1-2, pp.25-40, 2001.
  • Bourlard, H., Bengio, S., and  Weber, K., “New Aproaches Towards Robust and Adaptive Speech Recognition,” invited keynote, to be published in the Proceedings of Advanced Neural Information Processing Systems (NIPS13)} workshop, T.K. Leen, T.G. Dietterich, and V. Tresp (Eds.), MIT Press, Denver (CO), Nov.27-Dec.2, 2000.