BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Automatic Selection of Recognition Errors by Respeaking the Intend
 ed Text - Keith Vertanen (University of Cambridge)
DTSTART:20100208T110000Z
DTEND:20100208T120000Z
UID:TALK21856@talks.cam.ac.uk
CONTACT:Emli-Mari Nel
DESCRIPTION:We investigate how to automatically align spoken corrections w
 ith an initial speech recognition result. Such automatic alignment would e
 nable one-step voice-only correction in which users simply respeak their i
 ntended text. We present three new models for automatically aligning corre
 ctions: a 1-best model\, a word confusion network model\, and a revision m
 odel. The revision model allows users to alter what they intended to write
  even when the initial recognition was completely correct. We evaluate our
  models with data gathered from two user studies. We show that providing j
 ust a single correct word of context dramatically improves alignment succe
 ss from 65% to 84%. We find that a majority of users provide such context 
 without being explicitly instructed to do so. We find that the revision mo
 del is superior when users modify words in their initial recognition\, imp
 roving alignment success from 73% to 83%. We show how our models can easil
 y incorporate prior information about correction location and we show that
  such information aids alignment success. Last\, we observe that users spe
 ak their intended text faster and with fewer re-recordings than if they ar
 e forced to speak misrecognized text.\n\nThe paper is available here: http
 ://www.keithv.com/pub/autoselect/\n\n\n
LOCATION:TCM Seminar Room\, Cavendish Laboratory\, Department of Physics
END:VEVENT
END:VCALENDAR