Trainable grammar-based language modeling
At the heart of the Regulus system is a method for creating trainable
grammar-based language models for speech recognition. The major components of
this method are a Grammar Specialization technique, reusable general grammars
and the Regulus Grammar compiler that converts from Unification Grammars to
Context Free Grammars. This system supports creation of language models using
very small sets of examples.
Support for System Development
Regulus provides a development environment for developing spoken dialogue
system and speech translation systems. We have tools for testing, extensive
example applications and grammar resources.
Applications
Regulus has been used to build a number of non-trivial applications. The Clarrissa
Procedure Navigator was built to support Astronauts on the International
Space Station. Clarissa became the first spoken dialogue system in space when it
was tested on the ISS in 2005 by astronaut John Phillips. Our UCSC research
project with Ford Motors has built an extensive spoken dialogue system prototype
designed for in-car use. The Open Source MedSLT
Medical Speech translator covers source and target languages including English,
French, Japanese, Spanish, Catalan, Finnish and Arabic. Regulus has also been
used in projects by Fluency Voice, Gothenburg University, and Stony Brook
Univeristy.
Book
Our book about Regulus, Putting Linguistics into Speech Recognition, is published by CSLI Press. The book gives a detailed tutorial on Regulus and goes into the research behind it. It is available in a very resaonably priced paperback format.
Teaching
Regulus has been used twice as the basis for a very successful hands on spoken dialogue systems course at UCSC. Students construct spoken dialogue systems using Regulus for their course projects and give public demonstrations of their project at the end of the quarter. The paper we had at TeachCL 2008 gives more detail about the course.