Third European Conference on Speech Communication and Technology

Berlin, Germany
September 22-25, 1993


Using Anti-Grammar and Semantic Categories for the Recognition of Spontaneous Speech

Russell J. Collingham, Roberto Garigliano

Artificial Intelligence Systems Research Group, Computer Science (SECS), University of Durham, England, UK

This paper provides an introduction to the syntactic and semantic sub-system of AURAID: a speech recognition aid for use by deaf students in lectures. This sub-system produces a useful word recognition level from a continuous sequence of phonemes as could be provided by a continuous speech phoneme recognition system. The dynamic programming stage matches the phoneme input with a dictionary to produce a word lattice. The parsing stage makes use of an "anti-grammar" and semantic categories in order to determine the best sequence of words through the lattice. AURAID has a vocabulary of 2200 words and works in real-time using a simulated continuous speech phoneme recognition system (modelled on the performance of the DRA (UK) Speech Research Unit's Armada system). The phoneme error rate provided by this simulation is approximately 26%. Word recognition rates of approximately 85% have been achieved on sections of the simulated data using unrestricted speech. The simulated data is taken from real University lectures on the subject of software engineering.

Keywords: continuous speech recognition, spontaneous speech, grammar, semantic categories

Full Paper

Bibliographic reference.  Collingham, Russell J. / Garigliano, Roberto (1993): "Using anti-grammar and semantic categories for the recognition of spontaneous speech", In EUROSPEECH'93, 1951-1954.