Lily Clifford was partway through a PhD in computational linguistics at Stanford when she decided the more interesting problem wasn’t in her dissertation. It was in getting a computer to sound like it meant what it said. “There’s no unit test for sounding like you care,” she has said of the work — voice interaction, in her framing, is a design problem as much as an engineering one. She left the program to chase it.
She didn’t leave alone. Clifford recruited two people she already trusted with the problem: Brooke Larson, a fellow PhD linguist who was working as a language engineer on Amazon’s Alexa team, and Ares Geovanos, a Stanford-trained engineer who had been at UCSF building brain-computer interfaces for people who had lost the ability to speak. The three co-founded Rime in 2022, betting that the coming wave of enterprise AI would need voice models built for real conversation, not text-to-speech bolted onto a chatbot.
Building the dataset instead of buying the API
Most speech AI at the time was assembled the same way most language AI was: take a general-purpose model, scrape whatever audio you can license or find, fine-tune, ship. Rime’s founders — two linguists and a speech-interface engineer — took a slower route. Instead of scraping the web for training audio, the company built its own recording studio in San Francisco to capture the texture of real, imperfect, conversational speech: the pauses, the recoveries, the tone shifts that a script-read voiceover doesn’t have. It became, by the company’s account, one of the largest proprietary datasets of conversational speech built specifically for enterprise use.
That decision meant years without a funding announcement. Rime didn’t raise its first outside capital — a $5.5 million seed round led by Unusual Ventures, with participation from Founders You Should Know, Cadenza, and a group of angel investors — until May 2025, roughly three years after founding. For a category where competitors were closing much larger, much earlier rounds, that’s a long runway to spend heads-down on data and modeling before asking the market to price it.
The case for speech-to-speech
The bet paid off in the product Rime is now selling. Most voice AI stacks still work in three separate steps: speech-to-text, a language model to generate a response, text-to-speech to say it back — each stage adding latency and a chance to lose the tone of what was actually said. Rime has been building toward models that work directly in speech-to-speech, skipping the text bottleneck in the middle. The company describes it as building “the world’s first enterprise-ready speech-to-speech model” — software that responds to how something was said, not just the words a transcript captured, without the delay of translating audio to text and back again.
It’s a technically harder problem than chaining together existing APIs, but it’s the one that matters for the use case Rime has focused on: high-volume phone calls where a half-second of dead air or a badly-timed interruption reads as a machine, not a person. The company says its models now handle more than 100 million voice interactions a month across food service, healthcare, airlines, and fintech, with enterprise customers including Mayo Clinic, Dialpad, Upstart, and Asurion.
$24 million to keep the data lead
In July 2026, Rime closed a $24 million Series A led by M13 Ventures, with Twilio Ventures, Corazon Capital, and existing investor Unusual Ventures also participating — bringing the company’s total funding to roughly $30 million across the two rounds. “The design patterns that will define the speech-to-speech era of AI haven’t been built yet,” said M13 partner Morgan Blumberg, announcing the firm’s investment. “Rime is doing the foundational work by combining frontier AI with deep linguistic expertise.”
Rime says the new capital will go toward growing its roughly 35-person team — hiring more model developers and engineers — and building out enterprise customer-service, sales, and marketing-automation products on top of the core speech model. The company also recently added a chief science officer recruited from Meta’s Superintelligence Lab, a signal of how much of the Series A is aimed at research depth rather than just go-to-market spend.
The thesis holds up plainly
None of Rime’s pitch depends on a metaphor. A speech model that has to sound trustworthy on a healthcare intake call or a fintech support line has a narrower margin for error than one that just needs to sound impressive in a demo — and Clifford’s founding instinct, that this was a design problem before it was an engineering one, is the same instinct that led the company to spend three unfunded years building its own recording studio instead of shipping a faster, thinner product on someone else’s data. The market’s answer, for now, is a Series A that prices the slow build at $24 million and a customer list that includes some of the least forgiving industries in enterprise software.
Related founders
- Kyle Bigley Found a Faster Lane in SMS — another founder who bet on the two-way conversation, not the broadcast blast, as the differentiator in AI-powered customer messaging.
- The Tweet to $10M: Yasser Elsaid Built Chatbase With No VC and No Sales Team — a conversational-AI founder who took the opposite funding path, bootstrapping to eight figures in revenue before Rime had even raised its seed.
- How Dagne Dover’s Founders Chose Patient Capital Over VC — another founding team that spent years building before raising outside money, on the theory that the product needed to be right before the capital arrived.
The $24 million Series A, its investors (M13 Ventures leading, with Twilio Ventures, Corazon Capital, and Unusual Ventures participating), the 100 million monthly voice interactions figure, and the customer list (Mayo Clinic, Dialpad, Upstart, Asurion) are per Rime’s own announcement (rime.ai/resources/rime-series-a-announcement) and reporting from TechCrunch (“Rime picks up $24M Series A to help enterprises field customer calls,” July 15, 2026) and the Business Wire press release carrying the same figures. M13 partner Morgan Blumberg’s quote is per M13’s own announcement of the investment (unusual.vc and M13’s public statement on the round). The $5.5 million seed round, led by Unusual Ventures and announced in late May 2025, is per Axios (“Exclusive: AI voice startup Rime raises $5.5M seed,” May 29, 2025) and Rime’s own seed announcement (rime.ai/resources/seed). Lily Clifford’s departure from her Stanford computational-linguistics PhD, her “no unit test for sounding like you care” quote, and the company’s 2022 founding by Clifford, Brooke Larson, and Ares Geovanos, along with the co-founders’ backgrounds at Amazon Alexa and UCSF, are per reporting on Rime’s founding story and Clifford’s own public interviews. The chief science officer hire from Meta’s Superintelligence Lab is per public reporting on the Series A round.
