3 ex-Apple researchers want to make AI less awkward. Lightspeed, Accel, and Nvidia are giving them $50 million to do it.

Nuance cofounders all worked at Apple before launching their startup.

Nuance Nuance has raised $50M to build an audiovisual to audiovisual AI model.

Nuance wants to build AI chatbots with real-time reactions and human-like nuances.

Lightspeed led the round, with participation from Accel, Nvidia's NVentures, and South Park Commons.

A startup that's trying to make AI models better conversationalists with more emotional intelligence has just raised $50 million.

Seattle-based Nuance says it wants to build models that can hold face-to-face conversations without the lag and awkwardness of existing chatbots.

Fangchang Ma, Nuance's cofounder and CEO, said that existing voicebots or AI avatars are built by duct-taping together voice-to-text generation, a large language model responding to those texts, and text-to-voice generation.

This leads to a lot of latency and missing human nuances, such as reactions and microexpressions.

"If you use existing AI avatars, when they're listening, they're not reacting, or they're just doing random things," Ma said.

Ma, and his two co-founders, all of whom are former researchers at Apple, wanted to build a long-term solution, even though it would be more technically challenging.

"We decided to build this on one system, on one model.

There's audio video in and audio video out," Ma said.

Ma said that the data needed to train its models cannot be scraped from the internet and that it collects data in-house or from data partners.

"The perfect data for us will be, imagine the two of us having this conversation, we're having to record it," he said.