Title :
Noise-robust speech recognition by discriminative adaptation in parallel model combination
Author_Institution :
Fac. of Comput. & Electron. Eng., Keimyung Univ., Daegu, South Korea
fDate :
2/17/2000 12:00:00 AM
Abstract :
A discriminative adaptation method for parallel model combination (PMC) is proposed. For the adaptation, a modified version of PMC is adopted and the association factor defined in this variant is discriminatively adapted based on the generalised probabilistic descent (GPD) method. The proposed method is shown to be very effective at improving the recognition performances under extreme noise conditions with a small amount of adaptation data
Keywords :
adaptive signal detection; parameter estimation; speech recognition; adaptation data; association factor; discriminative adaptation; extreme noise conditions; generalised probabilistic descent; noise-robust speech recognition; parallel model combination; recognition performances;
Journal_Title :
Electronics Letters
DOI :
10.1049/el:20000277
