According to a report published on the TechCrunch website, San Francisco-based voice AI startup Rime has raised $24 million in a Series A funding round. The current round was led by venture capital firm M13 Ventures, with participation from existing investors including Twilio Ventures, Corazon Capital, and Unusual Ventures, alongside other existing investors who joined the round. Numerous voice AI startups are attempting to break into this market, where the most critical capability is handling calls for enterprises across sales, marketing, and customer support. Founded in 2022, Rime is trying to secure a competitive edge in the crowded market of AI-driven voice technologies by developing unique voice models trained on conversational data that it records independently, aiming to significantly reduce the customization burden required of its clients.
Company Foundation, Founders, and an Independent Recording Studio in San Francisco
The company was founded in 2022 by three partners: Lily Clifford, a former PhD student at Stanford University; Brooke Larson, a former engineer on Amazon's Alexa voice assistant team; and Ares Geovanos, an engineer from Stanford University. Instead of relying on collecting and scraping audio files and data from across the internet to train its voice models, the company decided to build its own dedicated recording studio in San Francisco. Through this studio, Rime independently collects and records the conversational spoken data used to develop its system. The company explained that it focuses on tuning its voice models to perfectly master the pronunciation of various client brand entities as well as specific terms characteristic of diverse industries.
Phonetic Architecture and the Technological Shift to Speech-to-Speech Models
To enable adaptation to different pronunciations without requiring clients to retrain models for the specific industry in which they operate, Rime employs a phoneme-based architecture. Early on, the company developed a technological infrastructure based on a pipeline of several separate models: a speech-to-text model, a text-to-speech model, and a large language model (LLM).
However, the company is now shifting its professional focus, directing its efforts toward developing more advanced speech-to-speech models. The primary objective of this transition is to significantly reduce system latency, improve the model's ability to manage natural turn-taking during a conversation, and better handle complex challenges such as varying background noises. Beyond this, the new technological approach aims to decrease the company's reliance on the orchestration of multiple separate models, thereby saving it from the need to manage a complex array of different models simultaneously.
The Company’s Stance on Current Limitations of Voice AI
Despite the major technological progress recently recorded in the development of voice AI technologies, Lily Clifford notes that large enterprises still prefer to use legacy Interactive Voice Response (IVR) systems, as AI-based voice technology still fails to match the level of effectiveness offered by those classic IVR systems.
According to Clifford, existing voice technology is not yet mature enough to automate the vast majority of phone calls in large organizations. She explained that while large language models (LLMs) have made the process of building functioning voice applications much simpler and easier, they have not fundamentally changed how the interaction itself feels for the end user. Clifford added that talking with a voice AI agent is currently not the most engaging or compelling experience for end users, describing it as a sort of new version of an IVR system, albeit one equipped with a better quality voice.
Team Expansion, the New Chief Scientist, and Client Roster
The company currently serves clients operating across a variety of sectors, including food service, healthcare, airlines, and financial technology (fintech). The company claims that thanks to its unique training data and the positioning of its models in the market, end users tend to stay longer on the phone line during calls, which has helped it secure enterprise contracts with large organizations, including Mayo Clinic, Dialpad, Upstart, and Asurion.
Following the current funding round, Rime plans to expand its current team of 35 people, aiming to hire additional professionals in model development, engineering, and business partnerships. The company recently brought on Rafael Valle to serve as Chief Scientist. Valle previously worked on audio understanding at Meta's Superintelligence Labs and on the applied deep learning and audio research team at NVIDIA. The current Series A round comes after the company previously raised $5.5 million in a seed round in May of last year.
The Investor Perspective and Market Competition
The market in which Rime operates is highly crowded and competitive. Large organizations currently tend to offload their call management to leading voice model developers like ElevenLabs and Deepgram; infrastructure companies in the sector such as Vapi, Retell, and LiveKit; or dedicated customer support companies such as Decagon and Sierra.
Morgan Blumberg of M13 Ventures, who is joining Rime's board of directors as part of the current funding round, shared his perspective in a conversation with TechCrunch. According to him, companies like ElevenLabs have transitioned to operating within the orchestration and application layer, thereby competing directly against the likes of Sierra and Decagon. Blumberg emphasized that, in his view, there is still much technical work to be done in this space, and that Rime's approach—focusing on pushing forward the best model characterized by low latency and exceptionally high reliability in a regulated environment—conspicuously sets it apart in the highly competitive market in which it operates.