COST 249: Continuous Speech Recognition over the Telephone

Project period: 94 05 - 98 04

Keywords: Continuous speech recognition, Markov models, neural networks, telephone services Project description

Project description
The main objective of the project is to co-ordinate research efforts in the area of multilingual continuous speech recognition for future public network services. This will be accomplished by establishing a unified language-independent speech recognition concept, and by investigating specific topics within this framework. This way it will be possible to validate the partners efforts in signal processing, statistical pattern recognition and linguistic processing in a unified way. One task will be to identify recognition subtasks, describe their internal communications, define control mechanisms for invoking them, and establish a methodology for evaluating their performance. Doing this it should be easy to exchange software modules, speech data and results. Another task is to propose a methodology for assessing the complexity of recognition tasks. This includes the development of tools for automatically assessing the variability in acoustic conditions, the size and confusability of vocabularies, the morphological structure of words, the perplexity of tasks, and the grammatical structure and semantic ambiguity of utterances. The tools will allow the research community to make fair comparisons of algorithms tested on different speech corpora in different languages. There will also be smaller tasks related to the investigation of specific topics covering focused aspects of the speech recognition problem. Source of funding

Source of funding: EU

COST 249 Work Group Organisation and List of Participants

COST 249 Homepage at Universiteit Gent