Overview
Approaches
- Path 1: HCI
- Use VR testbed to identify optimal perceptual perturbations for improving target speaker intelligibility.
- Objective measures of listening effort via EEG/fNIRS/pupil dilation
- Path 2: SP
- Use neuro-inspired approaches (oscillations, neuro-architectures, SSN, neural representations) for SOTA models
- Cross-modal fusion, using Gaze, MC Audio, and Video
- Combine
- Use outcomes of Path 2 to segregate target speaker speech
- Use outcomes of Path 1 to implement and evaluate speaker separation on smart glasses
Orga
- Anmeldung
- Früher anmelden, Bearbeitung kann sich ziehen
- Potentiell schon Ende April anmelden
- 2 Seiten Forschungsskizze
- Basierend auf CHAIS Proposal
- Allgemeiner halten als Forschungsprojekt
- Titel Richtung "speech enhancement in multi-speaker environments"
- Immatrikulation bei Uni Hamburg
- Erste paper üblich nach 6-12 Monaten (eher später)
- Erwartungen für Cum Diss
- Timo: 3 long paper (or 2 short for each long)
- Frank: 4-6 long paper
Conferences
SP
HCI
- ISMAR (AR)
- SIGCHI
- ACM SIGACCESS (accessibility!)
- UIST (interfaces)
- SIGCHI SUI (Spatial User Interaction)
- Mensch und Computer
Journals
SP
HCI
UKE
- Nature
- Nature Neurosci
- Current Biology
- Science Adv
- PNAS
- PLoS Biology
Resources