Rime picks up $24M Series A to help enterprises field customer calls

Voice AI has become one of the more crowded corners of the enterprise software world, with model makers like ElevenLabs and Deepgram, infrastructure players like Vapi and Retell, and support-focused startups like Decagon and Sierra all chasing the same prize: getting big companies to hand off their phone calls to software. Rime, based in San Francisco, thinks it has found a wedge. Rather than training its models on audio scraped from the internet, the company built its own recording studio and generates conversational data in-house, aiming to spare its clients from having to retrain models every time they need something to sound right.

That something, in Rime's case, is pronunciation. The company has leaned hard into getting brand names and industry jargon to come out correctly, using a phoneme-based architecture so customers don't have to fine-tune models themselves for their specific vertical. Rime was founded in 2022 by Lily Clifford, a former Stanford PhD student, alongside Brooke Larson, who previously worked on Amazon's Alexa team, and Stanford engineer Ares Geovanos. On Wednesday the company announced a $24 million Series A led by M13 Ventures, with Twilio Ventures, Corazon Capital, and Unusual Ventures returning as investors. That follows a $5.5 million seed round Rime raised last May.

Clifford's take on the state of voice AI is refreshingly unglamorous. She argues that despite all the progress, most enterprises still lean on old-school IVR systems because AI voice tech simply hasn't caught up. Large language models made it easier to build something that technically works, she said, but that hasn't changed how talking to one of these systems actually feels — it's still closer to a fancier IVR than a real conversation. That's a pointed admission from someone raising money in this space, and it frames why Rime is now shifting its engineering effort.

The company started out with the standard three-part pipeline: speech-to-text, a language model, then text-to-speech stitched together. Now it's pushing toward unified speech-to-speech models instead, chasing lower latency, smoother turn-taking, and better handling of background noise, while also cutting down on how many separate models it has to orchestrate. Rime says its approach keeps customers on the line longer, which it credits with helping land contracts with Mayo Clinic, Dialpad, Upstart, and Asurion across food service, healthcare, airlines, and fintech. The new funding will grow its 35-person team, with hiring aimed at model development, engineering, and partnerships. It also recently added Rafael Valle, formerly of Meta Superintelligence Labs and Nvidia's audio research group, as chief scientist. M13's Morgan Blumberg, who is joining Rime's board, framed the bet as a wager that there's still real technical ground to cover in voice AI, even as rivals like ElevenLabs push up into orchestration and application layers to compete directly with the Sierras and Decagons of the world.