Phonic, a Voice AI Platform, Secures Investment from Lux

AI-generated voices have gotten pretty darn good, you know? They're up to snuff for stuff like audiobooks, podcasts, reading articles out loud, and even basic customer support. But, a lot of businesses still aren't totally sold on the reliability of AI voice tech for their operations.

That's where Moin Nadeem and Nikhil Murthy, a couple of MIT grads, come in. They started Phonic, a company that's all about beefing up the reliability of synthetic voices while cutting down on the lag. These guys have been buddies for over seven years, ever since they met at MIT. When they kicked off Phonic last year, they noticed a gap in the market—no one was really offering a full-on voice tech solution.

"Voice AI is at a point where you're stitching together different bits, like automatic voice recognition and text-to-speech, and then you add some smarts," Murthy explained to TechCrunch. "But when we chatted with real customers, we realized there's a shortage of solutions that can handle things reliably on a large scale."

Nadeem, who used to work at MosaicML (which Databricks snapped up for a cool $1.3 billion in 2023), pointed out that a lot of companies in the voice AI space, like Vapi and Rounded, are just cobbling together different AI models. Phonic, on the other hand, does things differently—they train their models from start to finish, all in-house. Murthy reckons this approach has some big perks.

"When you own the models, you can really bake in some solid reliability features right into the models themselves," he said. "If you don't control that layer, you're just trying to glue together bits that don't really mesh well."

Plus, Murthy mentioned that Phonic's way of doing things lets them host and run their models in a cost-effective way. They train their models on all sorts of recordings, including accented and muffled speech, to make sure they're super robust.

Right now, Phonic's working with a select group of partners in the insurance and healthcare industries, but they're gearing up for a wider launch in a few months. Nadeem said soon enough, anyone interested can give Phonic's tech a whirl right from their website.

Phonic's managed to pull in $4 million in seed funding, with Lux leading the round and some big names like Replit co-founder Amjad Masad, Hugging Face co-founder Clem Delangue, Applied Intuition co-founder Qasar Younis, and Modal Labs founder Erik Bernhardsson chipping in.

Grace Isford from Lux Capital said they were drawn to Phonic's unique approach to training models in-house. "We think both Moin and Nikhil are incredible technologists," she said. "They started a machine learning club at MIT and have been working on training models for a while now. Their method of mixing diffusion and proprietary models in the voice AI sector is pretty innovative."