From Mexico — the interesting bit isn’t “another $50M for avatars.” Nuance Labs just closed a $50 million Series A to build one full-duplex audiovisual foundation model for face-to-face AI: words, tone, gaze, gesture, and timing in the same loop — not a stitched ASR/LLM/TTS/face pipeline. One contrast vs Gemini Live: Nuance is face + gesture + tone still in research preview later 2026; Gemini Live is a shipped speech-to-speech API.
SiliconANGLE and the company post match on the architecture: audio/video in and audio/video out on a single model that streams perception and response at the same time.
Why stitched ASR/LLM/TTS/face stacks feel awkward
Per Nuance’s release, today’s face-to-face systems usually transcribe speech, let an LLM decide the reply, synthesize a voice, then animate a face to match. Each handoff adds delay and drops the human layer — frozen faces while you talk, talking over you, random “listening” motion. CEO Fangchang Ma told SiliconANGLE the same duct-tape story: voice-to-text, language model, text-to-voice, then a fourth step for the face.
Nuance’s bet: one specialized full-duplex model that perceives the full range (words, tone, gaze, gesture, timing) and responds with facial and vocal expression in the same stream — showing it’s following while you’re still speaking.
$50M Series A — investors as sourced
Nuance (and SiliconANGLE) put the round at $50M Series A led by Lightspeed Venture Partners, with Accel, South Park Commons (SPC), NVIDIA, and Define Ventures among the participants. Founders: former Apple PhD researchers Fangchang Ma, Edward Zhang, and Karren Yang — Seattle-based. I’m not inventing ARR or customer lists.
Research preview later 2026 — not GA
Both sources say the company will open a public research preview later this year. No product GA claim. I’m not inventing latency bake-offs vs anything that already shipped.
My takeaway
From Mexico, the narrow read: Nuance Labs raised $50M (Lightspeed-led) to replace the stitched avatar stack with one full-duplex A/V model that treats conversation as simultaneous in/out — research preview later 2026, not a ship-today product. Sources: Nuance Labs, SiliconANGLE.
