All aspects of speech processing are rapidly growing in commercial importance. A leading market analyst reports 129.8 million USD were generated by speech products in North America in 2004, but predicts these revenues will increase to nearly one billion USD by 2011. In order to accelerate the rate of advance in speech technology, a new elite of researchers are required who are well-versed in all aspects of speech processing, as opposed to the narrowly-concentrated specialists who are currently the norm. Educating such an elite class of researchers will be a primary goal of the SCALE, Speech Communication with Adaptive Learning, Initial Training Network.
SCALE will be concerned with adaptive learning approaches to all areas of speech processing, with particular focuses on automatic speech recognition and synthesis, signal processing, human speech recognition, and machine learning. The primary goal of SCALE is to create an environment that will foster synergistic cooperation between researchers in each of the aforementioned disciplines. In particular, SCALE has three principal scientific objectives:
- To bridge the gap between speech recognition and speech synthesis;
- To bridge the gap between human and automatic speech recognition;
- To bridge the gap between signal processing and adaptive learning.
We plan to assess the improvements in these core technologies achieved in the course of the project through regular evaluations on standard benchmark tasks. We will also demonstrate the effectiveness of the interdiscplinary approach to research in speech processing proposed here through a series of high quality publications at international conferences and in prestigious journals. Finally, we propose to build cooperation between academia and industry through the inclusion of two major industrial partners in the consortium, whose presence will ensure the research training contributes directly to the next generation of speech products.