2013 Academic Year Seminars
Over the last forty years, remarkable progress has been made in the area of speech separation and enhancement, however accurate estimation of clean speech in real-world environments is still a challenge. We address the problem of speech estimation as statistical estimation with "missing" data in the independent component analysis (ICA) domain. Missing components are substituted by values drawn from "similar" data in a multi-faceted ICA representation of the complete data. I will present the algorithm for the inference of missing data in the case of a fixed pattern of missing data,and then I will present an application of our approach to the problem of bandwidth extension, where speech is degraded by a fixed filtering process. I will show the capability of the algorithm to reconstruct fine missing details of the original data with little artifacts: information of the source signal can indeed be modeled and used in order to recreate a natural sounding source in adverse conditions. The extension of the method to statistical spectral inference according to random patterns of missingness promises progress in the long open problem of performing speech enhancement while enhancing the intelligibility of speech.
This work has been done in collaboration with Doru-Cristian Balcan (Carnegie Mellon University) and Timo Gerkmann (Bochum University).
Siemens Corporate Research, Princeton, NJ
Justinian Rosca is Program Manager in Audio, Signal Processing and Wireless Communications at Siemens Corporate Research in Princeton, USA. He is also Affiliate Professor, Department of Electrical Engineering of University of Washington, Seattle, USA. He received the Dipl. Eng. degree in Computers and Control Engineering from Bucharest Polytechnic University in 1984, the M.S. and Ph.D. degrees in Computer Science from University of Rochester in 1992 and 1997 respectively.
Dr. Rosca is conducting research in signal processing and radio management, with an emphasis on topics involving acquisition, management and processing of data with uncertainties, such as statistical audio processing, blind signal separation, wireless management, adaptive principles in stochastic search and optimization, and probabilistic inference in artificial intelligence.
Dr. Rosca has more that two dozen US and international patents awarded, and more than 80 reviewed publications. He co-authored a book on solved problems in higher mathematics and co-edited the Proceedings of ICA 2006. He is presently on the editorial board of the Journal of Signal Processing Systems and the Journal for Genetic Programming and Evolvable Hardware both from Springer and serves as a member of committees of various conferences in the areas of machine learning and signal processing. Dr. Rosca chaired numerous sessions at international conferences, and events such the International Conference on Independent Component Analysis as program chair in 2006, and the Sparse Representations in Signal Processing workshop at Neural Information Processing Systems Conference in 2003. He gave multiple tutorials or invited talks on stochastic search techniques, audio signal processing and radio management at international and Siemens research events within the last ten years. Dr Rosca is a member of AAAI and IEEE.