For ElevenLabs, Synthesia and every other AI voice

Your AI voice can't say Kubernetes.
Now it can.

If you use ElevenLabs, Synthesia or any AI voice, you know the drill. Before every render, someone has to turn "3 March 2026" into "the third of March, twenty twenty-six", write "£4.2m" out in words, and try to stop it saying "Redis" wrong. Every script, by hand.

Paste it here instead, pick the tool you're using, and it comes back ready to go.

British English throughout. It already knows the names you'll be using, from Kubernetes and Azure to TD SYNNEX and Ingram Micro. Anything it doesn't know, tell it once and it remembers for your whole team.

Request access Sign in

A work email address is required. Personal mailboxes are not accepted.

What it does for you

Says the numbers properly

Dates, prices, percentages and decimals come back as the words a person would say. "47%" becomes "forty-seven percent" and "£12,500" becomes "twelve thousand five hundred pounds". It knows to leave things like "the 1990s", "S3" and "version 2.5.1" alone.

Gets the names right

Technical and brand names are marked up so the voice says them correctly, wherever your tool allows it. Where it doesn't, you get a short list of the names to set in its editor instead, so none of them catch you out.

Fits the tool you're using

Every voice tool takes different pause and pronunciation tags, and each has its own size limit. Pick yours and the output is written to match, so you don't get a tag read out loud as a word, or a script that stops halfway through.

Here's what that looks like

What went in

In today's rapidly evolving landscape, we are delving into the transformative power of Kubernetes. We saw a 47% uplift on 3 March 2026, cutting costs by £12,500 and moving 2.3TB to Cloudflare R2, with Redis in front.

What came back

We saw a forty-seven percent uplift on the third of March, twenty twenty-six, cutting costs by twelve thousand five hundred pounds and moving two point three TB to /klˈaʊd flˈeə/ R2, with /ɹˈɛdɪs/ in front.

The bits in slashes are how ElevenLabs is told to say a name. Paste them in exactly as they are.

01

Slop

The phrases that give away machine writing, like "in today's rapidly evolving landscape", are spotted and taken out.

02

Rewrite

What's left is tidied up, then checked. Every number, date, price and name still has to be there. If one has gone missing, that rewrite is thrown away and done again.

03

Numbers

Figures, dates, money and percentages become the words a person would actually say. The third of March, twenty twenty-six. Twelve thousand five hundred pounds.

04

Coverage

If you're cloning a voice, it checks which British English sounds your script never makes, and gives you a few extra lines to record so the clone has heard them all.

05

Pronunciation

Technical and brand names are looked up and marked so they're said correctly. Redis is reddiss, not re-deez. Your own names are saved and beat the dictionary.

06

Format

Pauses and pronunciation tags come out in whatever form your voice tool accepts, and the script is split up to fit its limit.

“

We built this for ourselves first. Every Compare the Cloud and Disruptive Live script used to go through the same hour of rewriting before it got anywhere near a voice. Now it's one paste.

CTC
Compare the Cloud
Built in London

Works with the tool you already use

Pick where the script is going and what comes back is written for it. Numbers and dates are spelled out whichever tool you use. Pronunciation marks go in where the tool accepts them, and where it doesn't, you get a list of the names to set yourself.

ToolPauses markedNames markedCharacters per request
Amazon Polly
Pauses and names both marked. 6,000 characters per request; you are billed for 3,000, and the tags are free.
 Yes SSML, IPA 6,000
Cartesia Sonic 3
Pauses marked and numbers written out. Cartesia takes no pronunciation marks, so names to watch are listed for you. No published limit; 2,000 is a safe chunk, and the Turbo variant caps at 500.
 Yes No 2,000
Deepgram Aura 2
Names marked in the form Deepgram reads, inside the text. Pauses come from your punctuation. Hard limit of 2,000 characters.
 No Inline JSON, IPA 2,000
ElevenLabs Flash v2.5
Pauses and names both marked, and the biggest limit here: a whole script in one go.
 Yes SSML, ARPAbet 40,000
ElevenLabs Multilingual v2
Pauses marked. This model takes no pronunciation marks, so tricky names are listed for you to check by ear.
 Yes No 10,000
ElevenLabs v3
Names marked inline between slashes, ready to paste. Pauses come from your punctuation.
 No Inline IPA 5,000
Google Cloud Text-to-Speech
Pauses and names both marked. 5,000 bytes per request, tags included.
 Yes SSML, IPA 5,000
Hume Octave 2
Numbers and dates written out. Hume takes no tags; tone goes in its separate description box, so names to watch are listed for you. 5,000 characters per utterance.
 No No 5,000
Inworld TTS
Numbers and dates written out, split into safe 2,000-character chunks. This tool has not confirmed it takes tags, so names to watch are listed for you.
 No No 2,000
LMNT
Numbers and dates written out, split into safe 2,000-character chunks. This tool has not confirmed it takes tags, so names to watch are listed for you.
 No No 2,000
Microsoft Azure AI Speech
Pauses and names both marked. Azure limits by minutes of audio rather than characters; 8,000 is a safe chunk.
 Yes SSML, IPA 8,000
MiniMax Audio
Numbers and dates written out, split into safe 2,000-character chunks. This tool has not confirmed it takes tags, so names to watch are listed for you.
 No No 2,000
Murf
Numbers and dates written out, split into safe 2,000-character chunks. This tool has not confirmed it takes tags, so names to watch are listed for you.
 No No 2,000
OpenAI gpt-4o-mini-tts
Numbers and dates written out. OpenAI takes no tags; tone goes in its separate instructions box, so names to watch are listed for you.
 No No 4,096
OpenAI tts-1 and tts-1-hd
Numbers and dates written out. These models take no tags and no instructions, so the words have to do all the work; names to watch are listed for you.
 No No 4,096
PlayAI
Numbers and dates written out, split into safe 2,000-character chunks. This tool has not confirmed it takes tags, so names to watch are listed for you.
 No No 2,000
Resemble AI
Numbers and dates written out, split into safe 2,000-character chunks. This tool has not confirmed it takes tags, so names to watch are listed for you.
 No No 2,000
Rime
Numbers and dates written out, split into safe 2,000-character chunks. This tool has not confirmed it takes tags, so names to watch are listed for you.
 No No 2,000
Speechify
Numbers and dates written out, split into safe 2,000-character chunks. This tool has not confirmed it takes tags, so names to watch are listed for you.
 No No 2,000
Synthesia Express-Voice
Numbers and dates written out, ready to paste. Synthesia sets pronunciations in its own editor, not the script, so the names to set there are listed for you.
 No No 2,000
Something else, plain text only
Plain text with the numbers written out. Use this for any tool not listed.
 No No 5,000
Say it once · Say it in British English · Say the numbers out loud · Say reddiss, not re-deez ·

Try it on your next script

Request access with your work email. Most people get straight in. The rest are read by a real person here, usually the same day.

Request access