Maddy - ADHD-powered mess Mar 4
GrapheneOS: Talks shit about Fairphone and their 'ancient' hardware on socials, not considering them as an option at all.

Also GrapheneOS: Partners with Motorola/Lenovo despite more devices in their lineup having relatively mediocre chipsets than not.
Show threadnytpuMar 4
@maddy Oh yeah, GrapheneOS, the OS that's fine officially shipping sandboxed proprietary Google Play Services but will not ship a screen reader because it's GPLv3!
Show threadSenilMar 4
@nytpu @maddy wait, the fuck? they don't ship a screen reader? they're OK with being inaccessible???
Show threadGrapheneOSMar 6
@senil @nytpu @maddy We have a fork of the open source TalkBack which is maintained by a blind member of our community and is included in GrapheneOS. We've developed our own text-to-speech implementation meeting our requirements for better usability, out-of-the-box functionality, security and acceptable licensing. Our text-to-speech app is currently available for early testing and will be included in GrapheneOS soon. We then need to add support to our Setup Wizard for integrating TalkBack.
Show threadGrapheneOSMar 6
@senil @nytpu @maddy We were in the process of investigating a bunch of open source TTS apps and saw none of them supported the required Direct Boot support for TalkBack so we filed it as a feature request with several. It wasn't done with the intention of bundling them but rather because we wanted the apps to be improved. We found 1 open source TTS app which almost met our requirements (SherpaTTS) but it's too slow to be comfortably used with TalkBack so we decided to make our own and did.
Show threadGrapheneOSMar 6
@senil @nytpu @maddy Our text-to-speech app project is at https://github.com/GrapheneOS/SpeechServices and we're actively working on polishing it up for inclusion in the OS as the default enabled TTS service. TTS services handle untrusted input from other apps on the OS combined with a lot of the data they handle being privacy sensitive. It's also crucial for it to perform very well especially for the TalkBack role. SherpaTTS provides high quality output but is much too slow, which was the reason to make this.
GitHub - GrapheneOS/SpeechServices: Fast, efficient, and high-quality text-to-speech for GrapheneOS using state-of-the-art models running completely on-device.

Fast, efficient, and high-quality text-to-speech for GrapheneOS using state-of-the-art models running completely on-device. - GrapheneOS/SpeechServices

GitHub
Show threadFamilyCyclist

@GrapheneOS

Using FUTO (based on openAI whisper) with local trained data without network access works fine for us...
Can be found @ fdroid

@senil @nytpu @maddy

Mar 7 at 5:47amWeb
Show threadGrapheneOSMar 7
@FamilyCyclist @senil @nytpu @maddy That's speed-to-text rather than text-to-speech. We're going to be providing both as fully open source implementations. FUTO keyboard is source available rather than open source and Whisper is a so-called open model rather than an open source model meaning the training data isn't available and it isn't possible to redo how they generated it. We want these to be open source and that's what we did for our existing text-to-speech app already.