Rime Launches Prism: The First Fine-Tunable Frontier Speech-to-Speech Model
Rime, the leading developer of enterprise voice AI models, led by Lily Clifford, Brooke Larson, and Ares Geovanos, has launched Prism, the first fully customizable frontier speech-to-speech (S2S) model. Brands can run Prism entirely within their own infrastructure while maintaining control over their data, conversation latency, vocabulary, pronunciation, and safety guardrails.
Enterprise Voice AI: Limitations of Rigid, Opaque Models.
While speech-to-speech (S2S) models deliver natural, fluid conversations, they often lack the emotional intelligence needed for effective human interaction. Their black-box architecture limits brands’ ability to fine-tune models to recognize subtle emotional cues such as frustration, hesitation, and confusion.
In high-stakes industries like healthcare and finance, missing these signals can escalate tensions, lead to abandoned calls, and increase reliance on human intervention, resulting in lost revenue and potential brand damage.
Rime's Solution: Prism for Customization and Control.
Rime has introduced Prism, a customizable, enterprise-grade speech-to-speech model designed to combine natural conversational quality with business control. Trained on millions of hours of human conversations recorded in Rime’s proprietary sound studios, Prism aims to improve call performance while reducing the need for human intervention.
Key Features:
Architectural Guardrails: Generates text before speech, enabling internal rules and policies to review responses before they reach customers.
Deployment Control: Supports on-premises and private-cloud deployment, with SOC 2 and HIPAA compliance.
Auditability: Text-first generation provides businesses with comprehensive, accessible audit trails.
Emotional Understanding: Recognizes customer emotions to respond more effectively and route calls appropriately.
Advanced Conversational Intelligence: Uses proprietary reinforcement-learning post-training on thousands of hours of human conversations to detect frustration and hesitation, reduce dead air, and improve call containment rates.
"A generic model with a custom prompt is not your voice. It’s a rental with your logo on it,” said Lily Clifford, Co-founder and CEO of Rime. “Prism is the first speech-to-speech model that becomes yours: it learns from your conversations, lets you set the latency and guardrails, and lives on your infrastructure. Quickly, the model running your calls will be one no competitor can buy.”
Prism Outperforms Major Consumer S2S Models.
In a real-world enterprise trial with Revmo, a fine-tuned version of Rime’s Prism outperformed OpenAI’s GPT across key business metrics.
Order confirmations and pickup methods: Increased from 68% with OpenAI to 100% with Prism.
Calls without hallucinations: Rose from 35% to 85%, a reported 142% improvement.
Calls without dead air: Improved from 66% to 88%, helping deliver smoother, more responsive customer conversations.
"Dialpad has been a design partner on Prism from the start. We're looking forward to rolling it out across our agentic products," said Shezan Kazi, VP of AI Products at Dialpad
World-Class Academic Founding Team.
Founded in 2022 by voice AI veterans from Stanford and Amazon Alexa, Rime has established itself as a proprietary AI modeling company. Instead of relying on scraped web data, the company built custom recording studios to capture thousands of hours of real, channel-separated human conversations, including natural interruptions and hesitations.
Rime’s models now automate millions of complex, high-retention calls for major organizations, including Asurion, Mayo Clinic, Upstart, and Dialpad.
