BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Speaker Retrieval in the Wild: Challenges\, Effectiveness and Robu
 stness - Erfan Loweimi\, Cambridge University Engineering Department
DTSTART:20240318T120000Z
DTEND:20240318T130000Z
UID:TALK213181@talks.cam.ac.uk
CONTACT:Simon Webster McKnight
DESCRIPTION:Effective speaker retrieval in real-world applications is an i
 mportant problem with extensive applications\, given the vastness of avail
 able media archives. In this talk\, we investigate the speaker retrieval s
 ystems developed by CUED in the context of the EPSRC-funded MVSE (Multimod
 al Video Search by Example) project. While we focus on the BBC Rewind corp
 us (1948-1979)\, our framework addresses the broader issue of speaker retr
 ieval on extensive and possibly aged archives. \n \nWe explore various cha
 llenges encountered in developing a speaker retrieval system in the wild\,
  addressing two primary issues: the dataset's unsuitability for direct tra
 ining and performance evaluation due to noisy and unreliable metadata\, an
 d the unconstrained acoustic conditions encountered in the archive\, rangi
 ng from quiet studios to adverse noisy real-world environments. \n \nVario
 us aspects of system development\, challenges\, potential solutions\, and 
 their functionality are examined\, along with systematic experiments condu
 cted in both clean setups and against various distortions to evaluate perf
 ormance. Additionally\, we touch on the utility of multimodal audio-visual
  speaker retrieval and analyse the synergy and consistency between these t
 wo modalities.
LOCATION:Zoom only: https://cam-ac-uk.zoom.us/j/86177109545?pwd=TmE1YlgzNW
 JKdGJQa1NQdk1kNS9zQT09
END:VEVENT
END:VCALENDAR