NA - New Paradigms for Speech Technology - Pilot

The current paradigm for speech modeling, in particular for automatic speech recognition (ASR) is based on statistical pattern recognition where model parameters are learnt from large collections of speech and text data. The performance of these systems a re, in spite of several decades of research, still well below human performance, and it is believed that automatic systems will never achieve human performance unless fundamentally new paradigms are investigated. In particular, it is believed that inter-d isciplinary approaches that combine utilization of existing and emerging knowledge about acoustics, speech, language, human perception and neural processing with statistical pattern recognition and decision theory have promises of being successful. Inve stigating fundamentally novel architectures will require a joint international effort in order to be successful. This pilot project aims at establishing a project plan for international research collaboration in this area. The planned project will primari ly involve Georgia Institute of Technology, the primary partner in a recently established US project, and NTNU, but will aim to be part of a larger, international effort, including leading research institutions in the USA, in the Nordic countries and in t he rest of Europe.