Rime Raises $24 Million to Improve Voice AI for Complex Calls

Rime has raised $24 million in Series A funding to build speech models for enterprises that field large volumes of customer calls. The round was led by M13, with participation from Twilio Ventures, Corazon Capital and existing investor Unusual Ventures. Chief executive and co-founder Lily Clifford announced the financing alongside the company's plans to push further into speech-to-speech systems designed for complex and regulated environments.

The company sits in one of the most crowded corners of applied artificial intelligence. Large organizations are offloading calls to voice model developers such as ElevenLabs and Deepgram, to infrastructure providers including Vapi, Retell and LiveKit, and to a growing number of vertical agent companies selling directly into contact centers. Rime's argument is that most of these systems fail in the same place: not at understanding the caller, but at sounding like something a caller is willing to keep talking to.

Trained on conversations, not audiobooks

Rime's technical distinction is its training data. Much of the industry's synthetic speech is trained on professionally recorded narration — audiobooks, voiceover libraries, studio sessions — which produces voices that are clean, articulate and subtly wrong for a phone call. Real conversation is full of hesitation, overlap, false starts, regional accent and the small tonal shifts that signal a person is listening rather than reciting.

The company trains on recorded conversational speech instead, and argues the result measurably changes caller behavior: fewer hang-ups, fewer demands to be transferred to a human, and longer completed interactions. In a channel where abandonment is the primary failure mode, that is the metric enterprises actually buy.

Why complex calls are the hard part

Simple calls were solved years ago. Automated systems can already confirm an appointment, read a balance or route a caller to the right department. The value that remains unautomated sits in calls that branch: a billing dispute that turns into a retention conversation, an insurance claim that requires collecting details in a specific compliance-mandated order, a healthcare intake where a mistake carries regulatory consequence.

Those calls demand more than transcription accuracy. The system has to hold context across several minutes, tolerate interruption, handle a caller who changes their mind halfway through, and hand off cleanly to a human when it reaches the edge of what it should decide. Latency matters enormously: a delay of even a few hundred milliseconds before a response reads as inattention, and callers start talking over the system.

Speech-to-speech architectures — where the model works directly on audio rather than converting to text, reasoning, and converting back — are the industry's current answer to that latency and expressiveness problem. Rime says the new capital funds exactly that work, alongside the enterprise controls that regulated buyers require before a synthetic voice is allowed to speak to their customers.

The investor case

The participation of Twilio Ventures is a useful signal. Twilio sits at the plumbing layer of enterprise communications and has unusually good visibility into what call volume actually looks like inside large organizations and where automation stalls. M13's lead reflects a broader thesis that voice, not chat, is the interface where AI reaches the largest number of ordinary customers, most of whom will never open a chatbot.

Competition remains the central risk. The largest model labs release improved speech capability on a regular cadence, and any startup whose advantage rests on voice quality alone is racing a curve that flattens quickly. Rime's defense is the combination of conversational training data, enterprise deployment controls and integrations that make replacement expensive once a system is embedded in a contact center's workflow.

For now, the market is expanding fast enough to support several winners. Enterprises that spent a decade routing calls to offshore centers are re-evaluating that entire cost structure, and the companies that can demonstrate a call handled end to end — without the caller asking for a human — are the ones being invited to the pilot.