Nuance Labs bags $50M in funding to fix the awkwardness of AI avatars
SiliconANGLE Mike Wheatley
Nuance Labs raised $50M to make AI avatars less awkward. It wants one model that listens, reacts, and talks back in real time, not a chain of laggy handoffs.
Based on reporting by SiliconANGLE, Mike Wheatley — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Seattle startup Nuance Labs has closed a $50 million Series A round to push ahead on a very specific fix for AI: making avatars feel less like polished puppets and more like something you can actually talk to. Lightspeed Venture Partners led the round. Accel and South Park Commons came back again, and Nvidia Corp. and Define Ventures joined in as new investors.
The company’s pitch is simple, at least on paper. Today’s voicebots usually stitch together several separate systems — one to turn speech into text, another to write the reply, then another to turn that reply back into audio. AI avatars add yet another step to animate a face. That stack creates delay, and it makes conversations feel stiff. Fangchang Ma, Nuance’s co-founder and chief executive, says the result is avatars that look frozen while someone else is talking.
Nuance wants to replace that relay race with a single “human foundation model.” The system it has built takes audio and video in, and sends audio and video out, using one full-duplex model instead of a bundle of parts. It watches for words, gaze, gestures, tone and timing, then responds in real time with facial and vocal expressions. Ma, who previously worked as an AI researcher at Apple Inc., says the model learns from conversations over time and can react while the user is still speaking.
The company has already posted a demo, though it’s still unfinished. Ma says the goal is more natural, more productive conversations for sales, customer service, coaching, professional training and education. He pointed to scenarios like a video call for language learning or interview practice. Nuance hasn’t launched a product yet, but it hopes to release its first research preview later this year.
The new money will speed up that work and help Nuance hire more researchers. Lightspeed’s Nnamdi Iregbulem called the opportunity “massive” and said the team had built a single model that follows users in real time. That may be the real bet here: not just better avatars, but a cleaner foundation for AI products that need to watch, listen and answer without looking like they’re thinking very hard about it.
My take — AI-written commentary, not fact-checked reporting
This is one of the few AI avatar pitches that sounds like an actual product problem, not a demo with nice lighting. The industry loves bolting three models together and calling it intelligence; Nuance is at least trying to stop the machine from tripping over its own shoelaces. Real-time interaction is where the clunk shows, and that’s exactly where the money is going.
Read more about this at: SiliconANGLE