BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:End-to-end contextual speech recognition with Tree-constrained poi
 nter generator - Brian Sun\, Cambridge University Engineering Department
DTSTART:20240219T120000Z
DTEND:20240219T130000Z
UID:TALK212269@talks.cam.ac.uk
CONTACT:Simon Webster McKnight
DESCRIPTION:Contextual knowledge is of vital importance to end-to-end auto
 matic speech recognition (ASR) systems\, especially for the long-tailed wo
 rd problem where systems suffer from degraded performance on rare or unsee
 n words that are both relevant to the context and carrying important infor
 mation. Integrating such contextual knowledge into such end-to-end systems
  is both necessary and challenging\, as contextual knowledge is always dyn
 amically changing while neural systems adopt a static set of trained param
 eters. In ASR\, dynamic contextual knowledge is often incorporated via con
 textual biasing\, where a list of rare words or phrases that are likely to
  appear in a given context is included\, denoted as a biasing list of bias
 ing words. A word is more likely to be correctly recognised if it is incor
 porated into the biasing list. This talk introduces tree-constrained point
 er generator (TCPGen) as an effective neural-based biasing component for e
 nd-to-end contextual ASR. TCPGen effectively integrate contextual knowledg
 e via a pointer generator mechanism\, and efficiently structures biasing l
 ists into prefix-trees. This talk includes the detailed TCPGen approach\, 
 the use of graph neural networks for tree encodings and its application to
  Whisper models.
LOCATION:Hybrid: JDB Teaching Room\, Engineering Department or Zoom: https
 ://cam-ac-uk.zoom.us/j/85153166972?pwd=bkgzN0lKejBjQ1BFSVVRbDAvbUdmUT09
END:VEVENT
END:VCALENDAR