How to import and exchange a macOS Personal Voice?

Question

macOS 14 (Sonoma) introduced Personal Voice, the ability to create a model of one's voice to be used with Live Speech accessibility feature (text-to-speech synthesis).

The training involves recording oneself speaking 150 separate sentences, which the software then chews on for six hours or so to generate a model of one's voice. This can then be used the way the built-in voices (Alex, etc.) are for Live Speech.

The feature has an export function that writes a folder containing .caf files of the 150 spoken sentences and a .json metadata file.

However, there's no corresponding import function, making the export currently pretty useless.

This is clearly a work-in-progress, but if someone is familiar with the internals I'd be grateful knowing:

• How to import the exported data to a new voice and tell the Mac to create a voice model?
• How to export and import a voice model?

Answer 1

Ok so I was facing the same thing, I wanted to know what to do with this export. First I want to say that it says it will share the voice across your iCloud devices so that relieved me some.

• I recorded a few phrases on my mac and then opened the dialog and told it to export the recordings.
• I opened the zip and got the file name.
• I did a search and found two files with that file name one in my export and one on the file system here: ~/Library/Group Containers/group.com.apple.accessibility.voicebanking/TTSVoiceBanking/Voices/<uuid>/*
• So I copied all my files from my export from when I recorded it on my phone and pasted the *.caf files into that folder above.
• I closed the personal voice window and opened it and it said "Recordings Complete"
• I am trying to create the voice now and I'm not sure how it will go... The SQL Lite databases I found in nearby directories did not have a straightforward clue.

Hope this sheds some light even if it isn't a solution.

Answer 2

(I don't know if giving a partial answer to my own question, but not accepting it, is legal, but here goes)

As of MacOS 14.1.1 (Sonoma), some functionality is clearly missing from the Personal Voice feature, and what's there is rather fragile. I found this procedure to work:

To import the set of .caf samples that was exported:

1. Using the System Settings > Accessibility > Personal Voice interface, click Create a Personal Voice.

2. Work your way through the dialogs for (a) naming the new voice, (b) checking the sound quality (ignoring any complaints about background noise), and (c) recording at least one sample sentence. Click Done to pause the recording session. This causes a new directory to be built: ~/Library/Group Containers/group.com.apple.accessibility.voicebanking/TTSVoiceBanking/Recordings/<uuid>/.

3. Copy into this directory the 150 .caf files that were exported. You needn't copy in the .json file.

4. Return to the Personal Voice interface and proceed as if resuming the recording session. The interface should notice that all 150 samples are now present, and begin building its model. For me, this took six hours.

5. When finished, the new voice should be available to Live Speech.

To use a Personal Voice from command line (e.g. say command in Terminal), you'll need to grant permission to your terminal app. Apple hasn't yet delivered this functionality in its Personal Voice UI so you'll need to use this relatively simple hack to do so.