AbbotDemo is a packaged demonstration of the Abbot system. AbbotDemo is designed to recognize British English and American English clearly spoken in a quiet acoustic environment. The demonstration system has a vocabulary of 10,000 words - anything spoken outside this vocabulary can not be recognised (and therefore will be recognised as another word or string of words). The vocabulary and grammar were optimised for the task of reading from a North American Business newspaper, for example the Wall Street Journal, although recent work frees this constraint somewhat.

The ABBOT system grew out of my PhD work on recurrent neural networks. It was further developed under the ESPRIT project "Auditory Connectionist Techniques for Speech" and then the ESPRIT project "WERNICKE: A Neural Network Based, Speaker Independent, Large Vocabulary, Continuous Speech Recognition System". Currently further development is being funded by the Framework 4 projects "SPRACH: Speech Recognition algorithms for connetionist hybrids" and "THISL: Thematic Indexing of Spoken Language".

The most comprehensive summary of the work is a book chapter that has now been published by Kluwer, but there are many other publications.

You can consult the README file for AbbotDemo or take a copy of the binaries.

A significant documentation effort is now underway.