The Pulse
Nuance Labs Raises $50M for Full-Duplex AI Avatars
Nuance Labs has raised $50 million in Series A funding to develop a full-duplex audiovisual model for face-to-face AI conversations. The Seattle startup says its system can perceive speech, tone, gaze and gesture while generating vocal and
AI.info Team ·
“The most productive collaboration comes from being able to express yourself freely, in words, tone, gesture, and expression, the way you would with a friend or close colleague.”
Fangchang Ma, co-founder and CEO, Nuance Labs
Nuance Labs announced on September 14 that it has raised $50 million in Series A funding to build AI avatars that can listen, interpret and respond during a conversation rather than waiting for each turn to finish. Lightspeed Venture Partners led the round, with returning investors Accel and South Park Commons and new investors NVIDIA and Define Ventures participating.
The Seattle company says its system is built around a single full-duplex audiovisual model. It takes in speech and visual signals while producing an audiovisual response at the same time, allowing an avatar to react while a person is still speaking. Nuance Labs says the approach is designed to address the delays, interruptions and blank expressions common in systems assembled from separate speech recognition, language, voice-generation and facial-animation models.
One Model for Listening and Responding
Nuance Labs describes its model as a foundation system for human conversation and expression. Rather than reducing an interaction to a transcript, the company says it analyzes words, tone, gaze, gesture and timing, then generates verbal and non-verbal responses that can include facial and vocal expression.
That distinction matters because conventional voice and avatar systems usually pass information through a sequence of separate components. Speech is transcribed, a language model generates an answer, a voice system speaks it, and an animation system attempts to synchronize a face with the output. Nuance Labs argues that each handoff adds delay and removes information about how a person is communicating.
Its model instead streams audiovisual perception into the system while streaming an audiovisual response out. The company says the architecture allows it to show that it is following a conversation through cues such as interjections, facial reactions and other signs of active listening.
Former Apple Researchers Target Face-to-Face AI
Nuance Labs was founded by Fangchang Ma, Edward Zhang and Karren Yang, whom the company identifies as former Apple PhD researchers. The founders worked on machine perception, reconstruction and audiovisual generation before starting the company, according to the funding announcement.
Nuance Labs says the team chose a single-model approach because existing systems were not designed to handle verbal and non-verbal communication together. The company’s stated goal is to create AI that can understand how people express themselves and respond in the moment, instead of forcing users to adjust their communication to the limits of a machine interface.
“They’ve never felt natural because we’ve been contorting ourselves and how we communicate to the machine rather than the machine adapting to us,” Ma said in the announcement.
Sales, Coaching and Education Among Target Uses
The company says the technology could support sales and customer service, coaching, professional training and education. Examples include AI systems that conduct interviews, help users practice a language or provide training through a live video interaction.
Nuance Labs has not announced a commercial product. It says the funding will support model development, expand its research team and prepare a public research preview later in 2026. The company is also hiring researchers and engineers across modeling, data, evaluation, inference and real-time serving.
Nnamdi Iregbulem, a partner at Lightspeed Venture Partners, said the company is addressing the interface between people and AI rather than building another general-purpose chatbot.
“There’s a massive opportunity to fix the way people interact with these transformative AI tools,” Iregbulem said. “Nuance Labs is building that interface, one that can see and respond to a person the way people do.”
The Funding Follows a Narrow Technical Bet
The $50 million round gives Nuance Labs capital to pursue a technically demanding product before it has released a public system. Real-time audiovisual interaction requires the model to interpret multiple streams, produce speech and visuals with low latency, and maintain a coherent response while the user continues to talk.
Nuance Labs frames that challenge as an architectural problem, not just an issue of adding more expressive animation to an existing chatbot. Its central claim is that perception and response should be trained and served as one continuous process. The public research preview will provide the first broader test of whether that design produces conversations that feel less interrupted and more responsive than current avatar systems.