HelixWorld: A Real-time Interactive Audio-Visual World Model
HelixWorld adds synchronized spatial stereo audio to an interactive world model running at 24 FPS on one GPU.
The paper says existing interactive world models are silent, while HelixWorld generates visual scenes and camera-grounded sound together under user control. The authors train it on a spatial audio-visual dataset with true stereo acoustics and metric camera poses, then distill it into a low-latency streaming student model. They also introduce HelixBench to test whether generated sound fields track viewpoint motion. HF Daily Papers' note
The paper says existing interactive world models are silent, while HelixWorld generates visual scenes and camera-grounded sound together under user control. The authors train it on a spatial audio-visual dataset with true stereo acoustics and metric camera poses, then distill it into a low-latency streaming student model. They also introduce HelixBench to test whether generated sound fields track viewpoint motion. HF Daily Papers' note
score 6