Rime raises $24 million to make AI phone calls feel less like talking to a robot

Key points

• Rime, a San Francisco voice AI startup, raised $24 million in a Series A round led by M13 Ventures in 2025.
• The company built its own recording studio to collect conversational training data rather than scraping audio from the web.
• Clients include Mayo Clinic, Dialpad, Upstart and Asurion across healthcare, airlines, food service and fintech.
• Rime previously raised $5.5 million in a seed round in May 2024.
• The company employs 35 people and plans to hire for model development, engineering and partnerships.

If you've ever called a company, hit zero seventeen times, and still ended up listening to a robot read a script at you, you already understand the problem Rime is trying to fix.

The startup, founded in 2022 by former Stanford PhD student Lily Clifford, ex-Amazon Alexa engineer Brooke Larson and Stanford engineer Ares Geovanos, announced a $24 million Series A this week. Lead investor M13 Ventures was joined by Twilio Ventures, Corazon Capital and Unusual Ventures, in a round first reported by TechCrunch AI.

The crowded field of companies building AI voice agents includes names like ElevenLabs, Deepgram and Vapi. Rime's angle is the training data. Instead of pulling audio from the internet, the team built a recording studio in San Francisco and recorded its own conversational material. The idea: voices trained on natural speech handle tricky brand names and industry jargon without the customer spending weeks teaching the model bespoke pronunciations.

Rime uses a phoneme-based architecture, meaning the system learns the building blocks of speech sounds rather than memorising whole words. That lets it adapt to different accents and terms without a full retrain every time a new client signs up.

Does any of this actually sound better?

Rime's own CEO says the whole category still has a long way to go. Clifford described today's AI voice agents as "kinda like a new IVR, but with a better voice", IVR being the interactive voice response system, the press-1-for-billing technology companies have used for decades. She told investors that most enterprises still trust those older systems because AI voice isn't yet reliable enough to handle the majority of calls.

That candour shapes where Rime is putting the new money. The company started with a pipeline of separate models: one to convert speech to text, one to generate a response, one to speak it back. It's now shifting to a speech-to-speech approach, where the AI listens and replies without converting everything to text in between. The goal is faster responses and more natural back-and-forth.

Rime claims its approach keeps callers on the line longer, which helped it land enterprise contracts with clients including Mayo Clinic and Asurion. The company's new Chief Scientist, Rafael Valle, previously worked on audio research at Meta Superintelligence Labs and NVIDIA. M13's Morgan Blumberg, who is joining Rime's board, told TechCrunch that the company's focus on low latency and reliability in regulated environments sets it apart from competitors moving into the application layer.

This is a space AI2Day has been watching: our coverage of Overtone, which uses AI voice for dating introductions, shows just how many bets are being placed on voice as the next interface people actually want to use.

On privacy: Rime records its own training audio, and its clients' call data flows through its systems. If your company uses Rime for customer calls, those conversations are processed by a third party. Check the small print if you work in a regulated industry.