ChatGPT text to speech: from an answer to an audio file
ChatGPT writes a video script, a story for the kids or a summary of a long article in seconds. The next step is often to hear it, or to turn it into a voice-over you can put in a video or a podcast.
This guide shows what ChatGPT can already read aloud, how to ask it for text that sounds good when spoken and how to voice that text with SuperClonica: natural and character voices in 31 languages, a free online test and a free engine that saves long scripts as audio files.
What ChatGPT can already do
When you need a recording you can keep, edit or publish, in a voice you choose, voice the text separately.
- The Read aloud button under an answer plays it in one of ChatGPT’s voices.
- Voice mode lets you talk with ChatGPT out loud.
- For developers, OpenAI sells a separate, paid text-to-speech API.
- For listening once, Read aloud is enough.
Voice a line free online
Copy a sentence of 30 to 100 characters from the answer, paste it into the page of any voice, choose the language and press Voice it. In a few seconds you can listen and download the recording, with no account.
Ask ChatGPT for text written for the ear
A prompt such as “Rewrite this for a voice-over: short sentences, numbers in words, no lists” does most of the work. Check the facts too: the voice reads mistakes just as confidently.
- Ask for short sentences and no lists, headings or bold text: marks such as # and ** are skipped and the structure is lost.
- Ask it to write numbers, prices and symbols as words: the engine skips signs such as % and $.
- Ask it to leave out links, since addresses are read in pieces.
- Give a length: about 90 words make 30 seconds of English speech at the default speed.
Long scripts as MP3
Paste the script into script.txt, install the free engine with python -m pip install superclonica-cpu and run superclonica --text-file script.txt --voice M3 --lang en --normalize --output script.mp3. The engine has no character limit and runs offline, so a ten-minute script becomes one MP3 without uploading anything. Change --seed for another take, or split the script into parts to give each its own voice.
Which voice to use
For videos you earn money with, use the ten stock voices, M1 to M5 and F1 to F5, free under the OpenRAIL-M license, or your own trained voice. Character voices from the catalog are fan-made and suit fan content and fun. Never present the voice as a real person, and say it is AI-generated where listeners would care.