A custom AI Speaker lets you generate audio in your own voice simply by typing. Use it to fix or adjust mistakes in your audio without re-recording, or to create text-to-speech from scratch in your projects.
This article covers:
- How to create a custom AI Speaker
- How to create an AI Speaker for a collaborator
- Consent and voice authorization requirements
Usage note
On current plans, this feature uses AI Credits. Learn more about tracking your Media Minutes and AI Credits.
Legacy and Sunset plans track usage differently. See our Understanding your Legacy and Sunset plan guide for details.
Create an AI Speaker
- Add a speaker label to your composition — click Add speaker (usually at the top of your composition).
- Select Create speaker from the dropdown, then name your speaker.
- Click your new speaker label, find your speaker name in the dropdown, hover over it, click the … menu, and choose Enable speech generation. This displays the consent and authorization script.
- Choose your microphone and click Record. The consent statement must be read in English, even if you’re creating non-English text-to-speech audio. For the best results, speak naturally, with varied tone and expression.
- Stop the recording, review your submission, and re-record if needed.
- When you’re satisfied, click Submit. You’ll receive a confirmation once your AI Speaker is ready (typically within minutes).
You can also create an AI Speaker from the AI Speakers tab in your Drive view.
Create an AI Speaker for a third-party
If your collaborator can’t record directly in the app, have them send you a recording of the consent statement. Complete steps 1–4 above, then:
For step 5, click the Choose a file button at the bottom of the Train and authorize speaker window. Select their recording file, then continue with steps 6 and 7.
Consent and voice authorization
Before Descript can create a custom AI Speaker, we require explicit recorded authorization from the person whose voice will be used (the “consenting speaker”). This training statement ensures the speaker clearly authorizes Descript to create an AI version of their voice, that the AI Speaker accurately matches the original voice, and that we can protect against unauthorized or fraudulent use.
Custom AI Speakers can only be created:
- By the consenting speaker themselves, or
- By someone acting on their behalf, with the consenting speaker’s recorded authorization.
Any attempt to bypass this process will be considered a breach of our Terms of Service.
We cannot authorize AI Speakers for:
- A deceased individual
- A non-consenting speaker
- Individuals unable to record the training and authorization statement
- Audio originating from an AI or artificial source
Related articles and workflows
If you run into issues with your AI Speaker, see our AI Speaker troubleshooting guide for tips on common problems and how to fix them.
You can also use your AI Speaker in these workflows:
- Regenerate audio — Replace or repair sections of audio without re-recording.
- Generate text-to-speech audio — Create new audio by typing directly into your script