How is Prism different from older voice AI?

Older systems convert speech to text and reply speech back to text, which can lose tone. Prism understands user speech directly but outputs its reply as text before speaking.

How fast does Prism respond?

It can react to a user's voice in about 300 milliseconds (0.3 seconds), and its first audio output took 0.57 seconds in testing.

Is Prism available to the public?

It is in trial use with some companies, and a general API is scheduled to be released in the future.

Audio & SpeechGIGAZINE AI4h agoOct 30, 2026, 22:00 JST

Rime's Prism voice AI shows replies for approval before speaking

Send on LINE

3 Key Points

What happened
Rime announced Prism, which reacts to speech in about 300 milliseconds, outputs its reply as text for review, and scored 56.6% on its τ-Voice test.
Why it matters
Companies can check what the AI will say, addressing a key trade-off in phone support: natural speech versus control over customer details and amounts spoken.

WHO IT HITSCustomer-service and contact-center teams that need AI to read customer details or amounts aloud accurately may be able to keep speech natural while reviewing replies in advance.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Get these free every morning

Context & Analysis

Voice AI has long faced a trade-off. Converting speech to text and back makes replies easy to check, but strips out anger, hesitation, and laughter. Understanding speech directly keeps those cues but makes it hard for companies to review what the AI will say. Prism's design aims to bridge the two: it understands the user's voice directly, yet prints its planned reply as text so companies can check and, if needed, fix it before it is spoken. It also keeps listening while speaking, so it can handle interruptions.

Rime's testing centered on τ-Voice, which simulates inquiries to airlines, retailers, and telecom carriers. Prism's 56.6% task success rate beat OpenAI and Google real-time voice models without extra reasoning by 17 to 30 points, and in telecom it reached 82.5% versus 45.6% for OpenAI's GPT-Live-1. Its first audio output came in 0.57 seconds, about half the time of GPT-Live-1 and Google's Gemini 3.8 Live.

Rime also envisions customizing Prism for each company's work. It is developing an API to train and tune Prism using conversation data with pilot companies, with the aim of letting businesses build voice AI specialized to their operations.

FAQ

How is Prism different from older voice AI?: Older systems convert speech to text and reply speech back to text, which can lose tone. Prism understands user speech directly but outputs its reply as text before speaking.
How fast does Prism respond?: It can react to a user's voice in about 300 milliseconds (0.3 seconds), and its first audio output took 0.57 seconds in testing.
Is Prism available to the public?: It is in trial use with some companies, and a general API is scheduled to be released in the future.

GIGAZINE AIRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.