Nuance Labs, a Seattle AI research lab founded by former Apple PhD researchers, announced a $50 million Series A on September 14, 2026, led by returning investor Lightspeed Venture Partners to advance a single foundation model for human conversation and expression.
Returning investors Accel and South Park Commons participated alongside new investors NVIDIA and Define Ventures. Lightspeed’s portfolio page records that the firm first invested in Nuance Labs at the seed stage in 2025, the year the company was founded, and lists partners Guru Chahal and Nnamdi Iregbulem on the company.
Nuance Labs said the funding will accelerate model development, expand its research team, and support its first public research preview. The company is hiring researchers and engineers across modeling, data, evaluation, inference, and real-time serving.
One Full-Duplex Model in Place of a Pipeline
The announcement argues that voice AI has grown while face-to-face AI has not, because existing avatar systems are pieced together from separate components. In the flow the company describes, speech is transcribed, a large language model decides what to say, a voice reads it, and a face is animated to match. Nuance Labs says those handoffs strip out the human layer of a conversation and add delay and choppiness, producing avatars that sit frozen while a person talks, talk over the user, or cannot keep up.
Nuance Labs says its model instead perceives and generates natively as one full-duplex system: audiovisual perception of the user streams in while an audiovisual response streams out at the same time, so the system can indicate through verbal and non-verbal cues that it is following while the person is still speaking. The company describes the model as perceiving words, tone, gaze, gesture, and timing; responding in real time with facial and vocal expression; and learning from how people actually behave in conversation, with real-time responsiveness built into the architecture rather than layered on top.
The company also contends that AI which sees only a transcript works from a partial picture of what the user wants, and that people say more, and say it more freely, when AI shows it is following them. The company’s About page characterizes the system as the first audiovisual model of its kind, one that sees, hears, reasons, speaks, and expresses in the same moment, in real time. Lightspeed describes Nuance Labs on its portfolio page as building a real-time human foundation model that brings social and emotional intelligence to voice, face, and body.
Founding Team and Targeted Uses
Nuance Labs was founded by Fangchang Ma, Edward Zhang, and Karren Yang and is headquartered in Seattle. According to the announcement, the founding team spent years at Apple working on how machines perceive and reconstruct people, where faithfully reconstructing a person for telepresence remained a challenge, let alone following the interplay of verbal and non-verbal cues in a live conversation. The founders started the company to understand and create human conversation and expression from audio and video signals natively in one model, rather than assembling transcription, language, speech, and animation components that were not built to work together.
The company’s About page lists Ma as cofounder and CEO, Zhang as cofounder and CTO, and Yang as cofounder and chief scientist, with Yaser Sheikh as an advisor. It describes the staff as research scientists and engineers who have published at top conferences and shipped ultra-low-latency machine-learning products to millions.
The release names sales and customer service, coaching, professional training, and education as settings where the company says expression drives results and AI works alongside people.
Executive and Investor Statements
In the Series A announcement, Ma said avatars have never felt natural because people have been contorting how they communicate to fit the machine rather than the machine adapting to them. “That’s what we’re building at Nuance Labs: AI that understands the many ways we express ourselves and responds the way a person does, in the moment,” he said.
Nnamdi Iregbulem, a partner at Lightspeed Venture Partners, said Nuance Labs is building an interface that can see and respond to a person the way people do. He credited the founding team’s combination of technical depth and operating experience, and said the firm sees the model becoming a core layer for AI products.
Nuance Labs plans to open the research preview to the public later this year, with access through a waitlist on its website.