Tsinghua's Falcon Extracts Target Speaker with 91.9% Fewer Parameters
Tsinghua and Qinghai University researchers released Falcon, a lightweight causal audio-visual target speaker extraction model that uses lip video to isolate a speaker's voice, achieving a 91.9% parameter reduction and 11x lower inference latency for latency-sensitive streaming applications.
2026-10-29 ~ 2026-10-30 · 2 related posts
- Falcon: lightweight streaming audio-visual speaker extraction with RealSSA benchmark — thu-ml · 2026-10-29
- Falcon: Streaming Audio-Visual Speaker Extraction Cuts Params 91.9% and Latency 11× — _akhaliq · 2026-10-30