This Is Apple's Personal Voice Accessibility Feature

Earlier this year, Apple teased a new accessibility feature called "Personal Voice", which allows you to create a synthesized voice of yourself if you are someone at risk of losing it completely due to a disability. Setting it up is done entirely on an Apple device; none of the recording samples you take are sent to Apple. Recording to train the feature takes about 15 minutes to do, but processing of your voice takes a lot longer (three days in my case). The feature will become generally available later this year, but already is quite promising.
The technology shown in this video is in beta and is not representative of the final product. Product features and capabilities subject to change. Some features may not be available on all devices nor in all areas. This feature should not yet be relied on in critical environments.
0:00 - Introduction
1:26 - Reading a sample script
2:48 - Personal Voice reading the same script
3:39 - Closing thoughts

[@sanderhoogeland9161]

Something like ElevenLabs is of course far better, but that is not on device. I definitely appreciate the accessibility efforts by Apple, as always.

[@pensane]

Very interesting. It captures the texture and register of your voice, but still sounds quite artificial. I don’t know if you use Spotify, but their AI DJ mode has a voice modeled after their CEO. It sounds very realistic. I’m guessing it involved an entirely different process… and more than 15 minutes. Thanks for sharing this. It’s a very cool feature.

[@FranceHopper]

Nope, I have Apple Music instead. Apple’s logic here isn’t too far from the process that historically has been used to make synthetic voices (like the original Siri voice), but it’s definitely a much more condensed process. What I’d like to see Apple do in the future is some way to capture additional samples, maybe automatically using (via opt-in) dictation samples. But given its intended audience, that might be difficult.

[@dironthomas9362]

it took a day to generate mine but I couldn't figure out to enable it until this video. I enabled it and it disappeared from my phone smh

[@FranceHopper]

I’m hoping it’s a feature that gets more stable in later builds. The generation process felt like the phone needed to be doing absolutely nothing else for it to process.