Research
Streaming audio-visual speaker extraction paper introduces RealSSA benchmark and Falcon model
Falcon is a causal method that cuts parameters by 91.9% and GPU latency 11.1 times versus the causal AV-TFGridNet baseline, and the RealSSA benchmark adds up to 15 far-field interfering speakers.
Every outlet that ran it 1 article
Hugging Face PapersResearchLead
Efficient Streaming Audio-Visual Target Speaker Extraction for Real-World Acoustic Scenes