The voice that no longer transcribes: Block and Rime process audio directly for call centers
Direct answer
The new generation of business voice processes audio directly, without converting it to text first. In the week of October 27, 2026, Block unveiled Dessa, a speech-to-speech model that listens and speaks at the same time, and Rime launched Prism, a speech-to-speech model for service calls that exposes the proposed reply for review before the caller hears it. The shared promise: preserving pauses, interruptions, laughter and tone, the information transcription drops along the way.
Block Dessa: conversation as training
- Announced October 23, 2026: Block, the company behind Square and Cash App, presented Dessa as a model trained only on human conversation audio, with no text and no pretrained model, per the company.
- Full-duplex: Dessa listens while it speaks, like two people on a call, instead of waiting its turn. Block says it learned from real conversations when to interrupt and when to signal it is following along.
- Faster than real time and runs locally: Block reported that inference runs faster than real time and that the model runs on a MacBook Pro M4 Max or better.
- Research preview, no public access: there is no public app, no API and no commercial launch date. Early access goes to a small group of AI safety researchers, and Block published no error rate, latency figure, model size or benchmark against competitors.
- The warning that comes with the announcement: human-sounding voices also lower the cost of phone fraud. PYMNTS and TechCrunch coverage collects concerns from filmmakers and journalists about deceptive calls, and the practical defense remains out-of-band verification: calling back on a known number or confirming inside the app.
Rime Prism: direct audio with a review point
- Launched October 28, 2026: Rime introduced Prism, a speech-to-speech model for customer-service calls, announced by CEO and co-founder Lily Clifford.
- Reads the caller's audio directly: Prism processes pauses, emotion and background sounds that a transcript misses, per the company.
- The reply is reviewed before speaking: the model exposes its proposed reply in text for review, a control point for companies that want direct-audio naturalness with human oversight before the caller hears the reply.
- 0.57 seconds to first audio: Rime reports that figure on its own benchmark, with no independent verification published as of launch.
- The buying question: natural-sounding voice and a correct answer are two different tests. Dessa demonstrated the first and has not published the second; Prism offers a review point so the company controls the second before it reaches the caller.
Note: this article summarizes Block's October 23, 2026 announcement and Rime's October 28, 2026 launch, plus PYMNTS, TechCrunch and RuntimeWire coverage. Performance figures are company-reported, with no independent verification published as of those dates, and Dessa remains a research preview with no public access.
Want AI that works for your business?
Message us on WhatsApp and let's talk about automation and AI agents for your business.
WhatsApp inicIARelated articles
- AI voice agent: what it is and how it answers calls for you
- Contact centers buy AI one job at a time
- What to do when your AI agent makes a mistake
Frequently asked questions
What changes when voice processes audio directly?
It preserves pauses, interruptions, laughter and tone that transcription drops, and reduces cascade errors between speech recognition, language model and speech synthesis.
Can a business use Block's Dessa today?
No. As of October 28, 2026 it remains a research preview, with early access only for AI safety researchers, no public API and no commercial date.
What does Rime's Prism offer that Dessa does not?
A review point: Prism exposes the proposed reply in text so the company can review it before the caller hears it, designed for service calls.
Does more human voice raise fraud risk?
Yes, per launch coverage: a convincing voice lowers the cost of phone deception, which is why out-of-band verification, such as calling back on a known number, gains importance.
