Physix Frontier · News Briefing Card (IT Home · Nov 19, 2026)
Doubao real-time audio-video call model 1.0 API goes fully live
KEY FACTS
- Volcano Engine has launched the API service for SeedRealtime 1.0, Doubao's real-time audio-video call model.
- The model unifies audio, video and temporal information into a single end-to-end model.
- The model can proactively speak up to remind users instead of only waiting for questions, and supports watching, listening and speaking simultaneously.
- According to the company, it cuts audio-video conversation rhythm issues by half compared with cascaded approaches.
- The model can distinguish multiple speakers, filter out background chatter, and resolve temporal references in the visual scene.
KEY DATA
50%Reduction in conversation rhythm issues
PHYSIX OBSERVATION
Replacing the ASR+VLM+TTS cascade with an end-to-end model eliminates stacked latency and information loss; speaking up proactively is the real differentiator. For developers, the interaction foundation for real-time voice assistants, smart hardware and companion apps is switching tracks, but the actual latency and stability in noisy environments after the full API rollout still need real-world testing.
Source: IT Home report
Physix Frontier