Title :
Porting: SwitchBoard to the VoiceMail task
Author :
Gales, M.J.F. ; Dong, Y. ; Povey, D. ; Woodland, P.C.
Author_Institution :
Dept. of Eng., Cambridge Univ., UK
Abstract :
The paper examines techniques that allow a well-trained source system built on one task to be rapidly adapted, or ported, to another target task. The two tasks considered are Hub5, or SwitchBoard, as the source system and VoiceMail as the target task. The two tasks are acoustically similar, both being telephone-bandwidth speech tasks, but differ in speaking style. SwitchBoard is conversational speech, VoiceMail is a set of voicemail messages. Various porting schemes for acoustic models are examined, including discriminative MAP and heteroscedastic LDA. Using around 28 hours of data, the error rate on VoiceMail was reduced by 42% relative compared to the baseline SwitchBoard performance.
Keywords :
learning (artificial intelligence); natural languages; speech recognition; Hub5; SwitchBoard; VoiceMail; conversational speech; discriminative MAP adaptation; error rate; heteroscedastic linear discriminant analysis; speech recognition; telephone-bandwidth speech tasks; well-trained source system; Bandwidth; Channel hot electron injection; Error analysis; Linear discriminant analysis; Maximum likelihood estimation; Mutual information; Speech recognition; Target recognition; Training data; Voice mail;
Conference_Titel :
Acoustics, Speech, and Signal Processing, 2003. Proceedings. (ICASSP '03). 2003 IEEE International Conference on
Print_ISBN :
0-7803-7663-3
DOI :
10.1109/ICASSP.2003.1198836
