How it works
From script to finished audio.
Exactly what happens when you press Generate, how credits are counted and which rules prepare your script for the voice.
The four steps
You write
Paste a script, add pauses, choose a voice, style, pace and format.
We prepare it
Your dictionary and smart reading rules rewrite the parts voices misread.
The voice reads it
The script is split at sentence boundaries, voiced, and joined with your pauses.
You download
MP3 or WAV plus an SRT subtitle file, in your library for 30 days.
How credits are counted
One credit is one character of the script you submit, including spaces and punctuation. Pause markers are not counted, and runs of spaces or line breaks count as a single space. A minute of narration is roughly 900 characters, so 10,000 free monthly credits are about 11 minutes of audio.
Credits are counted on your script, not on the rewritten text, so "$19.99" costs 6 credits even though it is read as "19 dollars and 99 cents". The studio shows the exact cost before you generate. Credits are reserved when you press Generate and returned automatically if the generation fails.
Smart reading rules
Smart reading is on by default and can be switched off per clip. It rewrites:
| You write | The voice reads |
|---|---|
$1,299 · €5 · £2.50 | 1,299 dollars · 5 euros · 2 pounds and 50 pence |
$19.99 · $5k · $2.5 million | 19 dollars and 99 cents · 5 thousand dollars · 2.5 million dollars |
15% | 15 percent |
2026-10-12 | October 12, 2026 |
(816) 555-0123 | 8 1 6, 5 5 5, 0 1 2 3 |
[email protected] | name at example dot com |
example.com/pricing | example dot com slash pricing |
e.g. · i.e. · etc. · vs. · 24/7 · & · #1 | for example · that is · et cetera · versus · twenty-four seven · and · number 1 |
Everything else is left as written. Dates in formats like 10/12/2026 are left alone because they mean different days in different countries; write them out or use 2026-10-12. Try the rules on your own text with the free TTS script cleaner.
Pronunciation dictionary
Add a word and how it should be said, using a simple respelling such as "Nguyen → win" or "Lumora → loo-MOR-uh". Entries match whole words without regard to case, are applied before smart reading, and work in the studio, bulk mode and the API. Our guide to fixing mispronunciations has respelling tips.
Pauses
Type [pause] for half a second of silence, or set a length such as [pause 1.5s] or [pause 800ms], up to 5 seconds. Pauses are exact silence inserted between the voiced parts. Commas, full stops and paragraph breaks also create natural pauses.
Formats and subtitles
- MP3: 44.1 kHz, 128 kbps mono. Small files for web, social and podcasts.
- WAV: 48 kHz, 16-bit mono PCM. The standard for video editors and broadcast. See MP3 vs WAV.
- SRT subtitles: one caption per sentence (long sentences are split), at most two lines of about 42 characters. Timings come from the length of each generated section of audio and are estimated within it, so they are accurate to the sentence, not the word.
Languages
The voices follow the language of your script, so a script in Spanish is read in Spanish. They are most natural in English; accents and fluency can vary in other languages. Smart reading rules are designed for English and can be switched off for other languages.
Files and retention
Generated audio files are kept for 30 days so you can download them again, then deleted automatically. Your scripts and clip history stay in your account until you delete them, so you can regenerate an expired clip at any time. Deleting a clip or your account removes its audio immediately.
What we don't do
- We don't offer voice cloning or imitate real people's voices.
- We don't allow impersonation, scams, robocalls or deceptive content. See the Acceptable Use Policy.
- We don't sell your scripts or audio, or use them to train AI models.
- We don't guarantee identical delivery between generations; like a human reading twice, it can vary slightly.
Hear your own script.
10,000 free credits every month (about 11 minutes of audio). No card required.