Skip to content

Cocktail Party Problem

First mentioned: https://pure.mpg.de/rest/items/item_2309493_5/component/file_2309492/content

Current survey: https://mdpi-res.com/BDCC/BDCC-09-00289/article_deploy/BDCC-09-00289.pdf?version=1763122319

Perceptual Aids

  • 6dB emphasize on focused speaker
    • Perceived halving of distance to speaker
    • Question: Emphasize via additive interference or resynthesis?
  • Real-time requirement
    • Look-ahead techniques (~5ms?)
  • Delay on switching focus?
    • Inertia to steering vector
  • Emphasize of real speaker/Attenuation of other speakers?
  • Increase speech brightness for better intelligibility (see #Meta Ray-Ban Glasses)
  • Increase spatial separation of speaker?
  • Increase pitch separation of speakers?
    • harder to differentiate similar pitched voices
    • simplify by maximizing pitch differentiation of target speaker
  • Minimize perceived distortion by combining minor perturbations along different dimensions?