9617484 Fanty, Mark A. Cole, Ron, A. Oregon Graduate Institute of Science and Technology CISE Research Instrumentation: File Server and Storage for Spoken Language Technologies This research instrumentation grant enables the purchase of a new file server and storage to be used for several research projects:- Segment-Based Language Recognition;Speaker Identification; Improving the Connectionist Approach to Automatize Speech Recognition;Robust and Portable Design Systems. Language recognition: the automatic identification of which language is being spoken from an excerpt of speech. A large 22 language telephone speech corpus is being collected. Speaker identification: the identification of who is talking from an excerpt of speech. A large corpus of realistic data collected over several channels across a large span of time is being collected. Improving the connectionist approach to automatic speech recognition: the extension of CSLU's neural network technology to large vocabulary recognition, providing an efficient alternative to the well-studied HMM systems. A large amount of data will be necessary. Robust and portable dialogue systems: the technology needed to build practical spoken language systems for the telephone packaged in a toolkit to make it accessible to a wide community. A large, regionally diverse cellular telephone corpus is being collected. Collectively, these corpora provide a valuable resource for research nation-wide as well as for the research described here. In each task domain, large amounts of annotated speech data are needed in order to study the sources of variability in speech, to train and evaluate recognition algorithms, and to compare research results across laboratories.
