home · mobile · calendar · colloquia · 2003-2004 · 

Colloquium - Pellom

The SONIC Large Vocabulary Speech Recognition System
Center for Speech and Language Research

The field of Automatic Speech Recognition (ASR) has a history dating back well over three decades. To-date machine recognition still lags human performance by one to two orders of magnitude. Most systems continue to fail miserably in the presence of noise, speaker accent, young and aging voices, or even conversational speech. For ASR to become a ubiquitous technology, significant research advances will need to be made in a number of key areas ranging from signal processing to data modeling and search.

To overcome these barriers, this talk will introduce SONIC, a large vocabulary continuous speech recognition platform developed at the University of Colorado. SONIC, which is freely distributed to other research labs for non-commercial use, currently achieves state-of-the-art performance on a number of commonly evaluated speech databases while also being computationally efficient, real-time, and rapidly portable to new languages. This talk presents an overview to SONIC and describes approaches used within the system to address major speech recognition problems through improved front-end signal processing, acoustic and language modeling, and search methods. This platform is used for teaching and training at the Center for Spoken Language Research (CSLR) and provides the basis for research within a wide variety of funded programs. The talk will conclude by demonstrating several ways in which this platform is being extended by other commercial and academic institutions and present a vision for the future.

Hosted by Dirk Grunwald.

Department of Computer Science
University of Colorado Boulder
Boulder, CO 80309-0430 USA
May 5, 2012 (14:13)