Output details
13 - Electrical and Electronic Engineering, Metallurgy and Materials
University of Birmingham : A - Electronic, Electrical and computer engineering
Language identification using multi-core processors
This paper describes the timely application of powerful graphics processor unit (GPU) technology to a major, international standard speech identification task. The time taken to process the 2003-US-NIST Language Identification task is dramatically reduced from 180 hours to 16 hours, significantly improving turn-around. This is also the first paper to compare alternative approaches to spectrum analysis for feature extraction from the dual perspectives of recognition performance and compatibility with the GPU architecture. This expertise in detection technology was key in securing £623k long-term funding from GCHQ, and is applied to stroke rehabilitation in the EU CogWatch project.