.svg)
.png)
20 Oct 26
Today we are launching Scaled Cognition Voice, the first text-to-speech model built specifically for enterprise customer experience. It delivers premium voice quality at the speed of a human conversation, and at $1 per million characters self-hosted, it is priced for the enterprises and BPOs that handle more customer calls than anyone else.
Enterprise CX teams no longer have to choose between the voice that sounds best and the voice they can afford to put on every call.
Made for the busiest lines in business
Most generative voice models were made for everything except a customer call. They narrate audiobooks and podcasts, voice video content and video games, read articles aloud, dub films, power smart speakers and consumer assistants, and perform character lines for creators. They were tuned to sound beautiful in a studio recording, where the text is clean, nobody interrupts, and no one is waiting on hold. They sound wonderful right up until a customer asks them to read back a confirmation code, a dollar amount or a street address, and the problem gets worse outside English. Voice was designed for the hardest audio environment there is, a live customer call over a phone line at contact-center scale.
Human-level latency. Voice starts speaking within 70 milliseconds of receiving the text, so the voice is never the reason a caller hears dead air. It adjusts pace and tone to the caller and handles interruptions without breaking stride.
It says it right the first time. Before it says a word, Voice works out what the sentence actually means, drawing on the same CX-tuned intelligence behind APT. It reads “Dr. Lee is on St. James St.” as “Doctor Lee is on Saint James Street,” and knows how to speak dates, addresses, account numbers and dollar amounts without being told. A custom lexicon lets your team control how it pronounces brand names and industry terms.
It speaks your customers’ language. Voice sounds native across languages and switches the moment your customer does, with no handoff to a different voice.
The millionth call sounds like the first. Tuned for phone audio, Voice keeps the same quality and speed whether it is handling a quiet afternoon or a holiday-weekend surge.
It sounds like your brand. Clone the voice your customers already know from a short recording, or design a new one.
Run it where your calls already live
Voice is available two ways. Self-hosted, it runs in your own cloud or data center for $1 per million characters. You run the infrastructure, we provide the model and support, and customer audio never leaves your environment. Placing Voice in the same region as your telephony and agents also keeps the round trip to the voice nearly instant. Hosted by Scaled Cognition, it is $9 per million characters with usage-based billing, so you can get to production without managing any infrastructure.
Start with Voice, scale to Super-Reliable AI Agents
Voice streams audio through a standard API, so anything that can send text and play back speech can use it, from modern AI agents and contact center platforms to legacy IVRs. Many enterprises still run IVRs with voices that sound robotic. Voice brings those systems into the generative era without rebuilding anything.
Paired with APT, our model built for super-reliable CX, Voice gets the context of the full conversation. APT handles the conversation and takes verified actions within business policy, and Voice speaks the result, reading confirmation codes and account numbers exactly as APT verified them.
Hear it for yourself
We would rather you judge Voice with your ears than take our word for it. Our blind listening test plays Voice alongside the best-known voice models without revealing which is which. Pick the one you would rather have talking to your customers. Early listeners have preferred Voice across the board, including against the most premium voices on the market.
Voice is generally available today. Visit the Voice product page to learn more, or schedule a demo to hear it on your own call flows.
ALSO BY
No items found.