MODULATION AND CHAOTIC ACOUSTIC FEATURES FOR SPEECH RECOGNITION * Dimitrios Dimitriadis † , Petros Maragos † , Vasilis Pitsikalis † and Alexandros Potamianos ‡ † Dept. ECE, National Technical University of Athens, Zografou, 15773 Athens, Greece ‡ Bell Laboratories, Lucent Technologies, 600 Mountain Ave., Murray Hill, NJ 07974, U.S.A. October 3, 2001 Abstract Automatic speech recognition systems can benefit from including into their acoustic process- ing part new features that account for various nonlinear and time-varying phenomena during speech production. In this paper, we develop robust methods for extracting novel acoustic features from speech signals based on nonlinear and time-varying models of speech. These new modulation- and chaotic-type features are integrated with the standard linear ones (mel- frequency cesptrum) to develop a generalized hybrid set of acoustic features. The efficacy by showing significant improvements in HMM-based phoneme recognition over the TIMIT database. Key Words: non-linear, modulation, chaotic, recognition, ASR. Accepted in Journal of Control and Intelligent Systems Special Issue on Nonlinear Speech Processing * This research work was supported by the Greek Secretariat for Research and Technology and by the European Union under the program EΠET-98 with Grant # 98ΓT26. It was also partially supported by the basic research program ARCHIMEDES of the NTUA Institute of Communication and Computer Systems. D. Dimitriadis, P. Maragos and V. Pitsikalis are with the National Technical University of Athens, Dept. of Electrical and Computer Engineering, Zografou, Athens 15773, Greece. E-mail: [ddim,maragos,vpitsik]@cs.ntua.gr. A. Potamianos is with Bell Laboratories, Lucent Technologies, 600 Mountain Ave., Murray Hill, NJ 07974, U.S.A. E-mail: potam@research.bell- labs.com 1