use case
verified July 2026
Narrate the whole paper, not just the headlines.
A newsroom produces an enormous number of words a day, across every section, and most of them never get read aloud. On a per-character meter the arithmetic is unforgiving, because the bill scales with total volume, so the rational move is to voice the top stories and let the rest stay silent. The reader hears a fraction of the edition and the archive is never touched.
01
Why the meter rations the edition
The volume is the whole point of a publication. A daily feed is hundreds of thousands of characters across news, business, culture and sport by definition, so a per-character price charges for the exact thing a newsroom cannot make smaller. The result is a triage nobody wants: narrate the front page, skip the section desks, leave the back catalogue mute.
A flat seat removes the triage. An article is a loop over requests and a full edition is a longer loop, and the loop costs the same whatever the day brings, so voicing every section every morning becomes an editorial decision rather than a budget one.
02
What the feed needs from the engine
- First audio at about 107 ms with streaming output, so a story is narrated and published the moment it clears the desk rather than waiting in a render queue.
- One cloned masthead voice carried across 23 languages, so every section and every localised edition speaks in the same identity.
- A pronunciation dictionary fixes proper nouns, place names and tickers once per request, so a byline or a company is read correctly across the whole feed.
- Break tags place the beat between the standfirst and the body, so the audio keeps the shape of the article instead of running the headline into the first sentence.
03
Notes — an engineer's checklist
01Can we narrate the entire daily feed rather than a selection?
Yes. Each request carries up to 2,000 characters and streams gap-free, so an article is a loop and the whole edition is a longer loop. On a flat seat the length of the day does not change the cost, so there is no reason to narrate only the top stories.
02How do we keep one house voice across sections and languages?
The voice is a request parameter, so every article is conditioned on the same reference clip and there is no drift between the business desk and the sport desk, or between the English and the localised editions across 23 languages.
See also — related sheets
use case
2,000
A chapter costs the same as a sentence
Long-form narration on an unmetered seat: a whole book is a loop over requests, and the length of the book stops being a line item on the invoice.
use case
24 kHz
A reading voice that is not a punishment
Accessible audio on an unmetered seat: read the whole document, every time, without the character count arguing for a shorter description.
use case
$0
A narrator who never gets tired of take nine
Script-to-voiceover on a flat seat: re-render the whole video after an edit, in your own cloned voice, without a character count deciding how long the script runs.
capability
0
Voice cloning from ten seconds, in the request
Zero-shot voice cloning: a ten-second reference rides inside each request, no training job, and the identity holds across 23 languages.
A key and one seat to build this on — the same production API this page measures.
Request access for this use case