Falcon: lightweight streaming audio-visual speaker extraction with RealSSA benchmark
thu-ml · hf · 2026-10-29
Tsinghua and Qinghai University researchers released Falcon, a lightweight causal audio-visual target speaker extraction (AVTSE) method delivering high quality with drastically reduced parameters and latency for latency-sensitive streaming applications.
- Paired with RealSSA, a new benchmark addressing gaps in existing ones that fix speaker count at two and ignore near/far-field interference, reverberation, and sudden noise
- Paper on arXiv, Oct 23, 2026
More from Multimodal
- ComfyUI launches Creative Partners Program with six studios already shipping client work — paultrillo · 2026-10-29
- Will Smith eating spaghetti: the AI video meme that won't die — Busy_Connection_8027 · 2026-10-29
- The heat equation is Gaussian blur: how 19th-century math underpins computer vision — techNmak · 2026-10-29
- Suno employee teases a 'big' product launch: 'the most fun thing I've worked on' — enggirlfriend · 2026-10-29
- Popular AI thread: open-source desktop Malfoy companion app built with Jimeng AI short-drama studio — op7418 · 2026-10-29
- Sunra launches aggregator API serving top image/video models at makers' own prices — EXM7777 · 2026-10-29