Rime Raises $24M to Build Voice AI on Audio It Records Itself

Instead of scraping the web for speech, the San Francisco startup runs its own recording studio — a data strategy its CEO argues is the only way to fix voice agents that still feel like an IVR with a nicer voice.

Jul 16, 2026Source: TechCrunch

Rime Raises $24M to Build Voice AI on Audio It Records Itself

Most voice AI companies train on whatever audio they can scrape. Rime built a recording studio. On July 15 the San Francisco startup announced a $24 million Series A led by M13 Ventures, with Twilio Ventures, Corazon Capital, Unusual Ventures and other existing investors joining. M13 partner Morgan Blumberg takes a board seat.

The founders

Rime was founded in 2022 by CEO Lily Clifford, who left a Stanford linguistics PhD to start it; Brooke Larson, a former Amazon Alexa engineer; and Stanford engineer Ares Geovanos. The company has 35 employees and is hiring across model development, engineering and partnerships. It recently added Rafael Valle as chief scientist, previously working on audio understanding at Meta Superintelligence Labs and on Nvidia's applied deep learning audio research team.

The data bet

The differentiator is provenance. Rime built a recording studio in San Francisco to collect its own conversational data rather than relying on scraped web audio — a choice that looks increasingly shrewd as scraping-based rivals face copyright scrutiny. It pairs that with a phoneme-based architecture designed to pronounce brand names and industry-specific terms correctly, so customers do not have to retrain a model for every vertical.

What is broken today

Clifford's diagnosis is unusually blunt for a founder raising money on the category. "The voice technology is still not there to automate the vast majority of enterprise phone calls," she told TechCrunch. "LLMs have made it a lot easier to build voice applications that work, but they haven't changed how it feels to interact. Talking with a voice AI agent is not the most compelling experience for the end user. It's kinda like a new IVR, but with a better voice." Her fix is architectural: moving off a stitched speech-to-text to LLM to text-to-speech pipeline toward speech-to-speech models that cut latency, improve turn-taking, handle background noise and reduce orchestration overhead.

Traction and the field

Customers include Mayo Clinic, Dialpad, Upstart and Asurion across food service, healthcare, airlines and fintech, and TechCrunch reports Rime handles over 100 million calls each month. The company competes with model developers ElevenLabs and Deepgram, infrastructure players Vapi, Retell and LiveKit, and customer-support firms Decagon and Sierra. Blumberg's thesis is that the model layer is still contested: "Companies like ElevenLabs have moved into being an orchestration and the application layer... I think there's just so much more to be done technically."

Why It Matters

Voice is the last interface where AI has to pass in real time, with no chance to reread the output, which is why the category keeps producing demos that impress and deployments that annoy. Rime's studio is a bet that the ceiling is data quality rather than model scale — expensive to build and hard to copy if it works. The strategic question is whether owning clean, licensed speech data becomes a moat as the copyright fights that Suno is now living through spread to voice.

Filed underAI Agents AI Tools #AI funding #M13 #Rime #speech models #voice AI