Voice AI startups’ biggest unlock has been handling calls for enterprises in areas like sales, marketing and customer support. Large organizations are offloading calls to voice model developers like ElevenLabs and Deepgram; infrastructure companies like Vapi, Retell, and LiveKit; and dedicated customer support shops like Decagon and Sierra.
San Francisco-based Rime is trying to gain an edge in this crowded market with its voice AI models that are trained on conversational data that it records, aiming to reduce its clients’ customization load.
Founded in 2022 by former Stanford PhD student Lily Clifford, ex-Amazon Alexa engineer Brooke Larson, and Stanford engineer Ares Geovanos, Rime has built a recording studio in San Francisco to collect its own conversational data rather than relying on scraping the web for audio.
The startup said it focuses on tuning its voice models to nail the pronunciation of different brand entities and industry-specific terms. It employs a phoneme-based architecture to adapt to different pronunciations so that customers don’t have to retrain models for their specific industry.
Rime on Wednesday said it has raised $24 million in a Series A funding round that was led by M13 Ventures. Twilio Ventures, Corazon Capital, Unusual Ventures and other existing investors also participated.
Clifford said that despite progress in voice AI development, enterprises still prefer legacy IVR implementations, as AI voice technology still can’t match up to IVR’s effectiveness.
“The voice technology is still not there to automate the vast majority of enterprise phone calls. 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,” she said.
The startup started off with a pipeline of separate models for speech-to-text, text-to-speech, and a large language model. But it is now shifting focus to develop better speech-to-speech models to reduce latency, improve turn-taking, and tackle issues like background noise. The new approach will also serve to decrease reliance on orchestration, so the company doesn’t have to manage a bun …