funding · September 14, 2026
By Joe Jurczyk, Editor
Nuance Labs raises $50M Series A for a full-duplex audiovisual conversation model
Seattle-based Nuance Labs, founded by three former Apple researchers, raised a $50M Series A led by Lightspeed with participation from NVIDIA. The company is building a single model that takes audio and video in and produces audio and video out, rather than chaining speech-to-text, an LLM and text-to-speech.
Nuance Labs announced a $50 million Series A on 14 September 2026. Lightspeed Venture Partners led, with returning investors Accel and South Park Commons and new investors NVIDIA and Define Ventures.
The company was founded by Fangchang Ma (CEO), Edward Zhang and Karren Yang, all former Apple PhD researchers who worked on machine perception and human reconstruction. Its stated approach is a full-duplex foundation model that streams audiovisual perception from the user while simultaneously generating a response with facial and vocal expression, taking in words, tone, gaze, gesture and timing. Nuance Labs says it will open a research preview to the public later in 2026, with a waitlist open now.
Why it matters for digital humans: Almost every conversational avatar in production today is a pipeline — ASR, then an LLM, then TTS, then a lip-sync or rendering layer — and the seams in that pipeline are what produce the stiff turn-taking, missed interruptions and dead air that practitioners spend their time papering over. A single model trained end to end on audio and video is the architectural answer to that problem, and it is the same bet several labs are now placing. Nothing here is verifiable yet: the company has published no latency figures, no benchmarks and no demo, and the release describes intent rather than results. NVIDIA's participation is worth noting given its own ACE stack. The research preview, when it lands, is the thing to judge.