For anyone with a script and no narrator
Turn a written script into AI speech
A voice generator reads exactly what it is given, so most of a good result is decided before the audio exists. This site covers the whole job: preparing the text, fixing pronunciation and pacing, choosing a voice and finishing the audio.
Cues for the writer
Write the script for the ear
A sentence that reads well can still sound wrong. These are the places a synthetic voice most often needs help from the page.
-
Cue: names
Respell what it could misread
Siobhan, written as shi-VAWN
A surname, a place or a product name can come out wrong. Respelling it by sound usually gets closer, and some tools keep a pronunciation list so the fix only has to be made once.
-
Cue: figures
Write numbers the way they are said
Not a string of digits but twelve hundred
A date, a total or a phone number can be read aloud in more than one way. Writing it out as the words you want removes the guess.
-
Cue: breath
Let punctuation set the pace
Stop here. Then carry on.
Voices tend to pause at commas and full stops, so a long sentence without them runs out of air. Split it where a person would take a breath.
-
Cue: stress
Make the key word land
It was not the first draft.
A voice can lean on the wrong word and change the meaning. Rewording so the key word arrives late in the sentence often helps, and some tools offer emphasis controls for the rest.
The running order
From a finished draft to finished audio
- First
Prepare the text
Take out what only makes sense to the eye: headings, links, footnotes, tables and anything in brackets. If a list matters, turn it into sentences.
- Next
Fix the pronunciation
Generate a short passage holding the hardest words before committing the whole script, and correct them while corrections are cheap.
- Then
Set the pacing
Adjust speed and pauses against how the audio will be used. A training module can wait for the listener in a way an advert cannot.
- After that
Choose the voice
Let the script decide its register, accent and energy, and try a voice on your own lines rather than a demo. Check the licence terms if the audio will be published or sold.
- Last
Finish the audio
Listen all the way through, regenerate only the lines that miss, then trim, level and export in the format the project needs.
As written, as spoken
Shorthand a reader skips and a voice has to say
- Dr. Lee
- Doctor Lee
- St. James St.
- Saint James Street
- approx.
- about
- e.g.
- for example
- Q3
- the third quarter
- R&D
- research and development
Kinds of script
What each kind of script asks of a voice
Narration and explainers
Long stretches of one voice, so what matters is that it stays comfortable to listen to over time.
Adverts and promos
Short and tightly timed, with energy that has to fit the slot and fall on the right words.
Training and e-learning
Clear and steady, with terms said the same way every time and scripts that get updated later.
Dialogue and characters
More than one voice in the same scene, which have to sound distinct and still belong together.
Articles on each step of the running order are on the way.
Who we areWhat readers ask
The questions this site is built around
- How much better is AI text to speech than the built in system voices?
- Which text to speech generator gives the most control over speed and emphasis?
- Which text to speech AI handles punctuation and pauses most naturally?
- Where can I find free text to audio conversion for a nonprofit newsletter?
- Can you recommend free text to speech software for making study materials accessible?
- Which text to speech tools sound closest to a real human narrator?
- Are there text to speech free options that allow commercial use?
- What AI voice text to speech platforms support multiple speakers in one script?
- What text to speech online services work without creating an account?
- Is there a text to audio free tool for converting long documents?