Open-source voice model listens continuously, decides every 0.4s whether to speak

Audio Interaction processes speech in real-time without waiting for recording to end, unlike GPT-4o or Qwen3.5-Omni

Published 9sem1 sourceNotable

≈ 22sSpeed

The fact

Model transcribes, translates, responds and detects ambient sounds (coughing, etc.) in a single continuous stream

Code, model weights and training data available open-source under Apache 2.0 license on GitHub

Click the link to read an article on the topic:

Explore this topic

Auto-synthesis from 1 media source · identified on June 6, 2026