Ex-Apple researchers raise $50M to make AI conversations less awkward
What's the deal? NuanceDealroom has a profile for this one. Try Dealroom →, a Seattle-based startup building AI models for face-to-face conversation, has raised a $50 million Series A led by Lightspeed Venture Partners. Accel, Nvidia's NVenturesDealroom has a profile for this one. Try Dealroom →, South Park Commons, and Define VenturesDealroom has a profile for this one. Try Dealroom → also participated. The round follows a $10 million seed led by Accel in July last year.
What's the endgame? Founded last year by three former Apple researchers, Nuance wants to build models that hold video conversations without the lag of existing chatbots. Chief executive officer Fangchang Ma said current voicebots and AI avatars are "duct-taping together" voice-to-text, a language model, and text-to-voice — creating latency and missing human cues like reactions and microexpressions.
How is it different? Rather than stitch tools together, Nuance is building a single model with audio and video in, and audio and video out. "If you use existing AI avatars, when they're listening, they're not reacting, or they're just doing random things," Ma said.
What will it build? The pre-product startup envisions users hopping on a video call with an AI to practice interviews, give presentations, or learn a language. It also plans to sell to businesses for AI interviewers, sales reps, and customer support. A public research preview is due later this year.
What could go wrong? The data Nuance needs cannot be scraped from the internet, Ma said, so it collects recordings in-house or through data partners. His ideal: clean, dual-sided audiovisual footage of natural conversations — "not performative."
Why now? Ma said the cash is funding the eight-person team, with hiring focused on research and development. The next step is bringing on go-to-market experts to set pricing and monetization.
The signal: Backing from Lightspeed, Accel, and Nvidia's venture arm signals investor appetite for AI that feels more human — and a bet that a single unified model can outperform the patchwork approach that dominates today.
Image credit: Generated with Gemini
