Arlas I tested the VOSK-Powered-Live-Subtitle-V3 app before I found this post, it is more like a (poor) substitute for Live Transcribe which serves a different purpose. Live Transcribe uses microphones ( the onboard microphones of the phone, Bluetooth, & USB mics) as the audio source. Live Captions uses System Audio as the audio source. That's a big difference.
Additionally, we can use the original Live Transcribe app on Graphene OS no problem. Also, Live Transcribe itself has been open-sourced, so, that would be a much better starting point with which to build an app for Pixel Phones 6 (not sure which variant) and newer.
On a side note, I was able to get the Google Recorder app working which depends on the Voice Synthesis app which also needs to be installed. This is great when we want a recording and a transcript after the session. If anyone installs this a tip for you, restart the app after getting the permissions sorted out and after installing the Voice Synthesis app and giving that the proper permissions.
I tried finding an audio routing and virtual audio device app that can tap system audio thinking that could be a workaround, I did not see anything.
I did find an APK package with Live Captions on APK Mirror, but I have not been able to figure out how to launch it. I'm not even sure if it works on non-Motorola phones. On stock Android, Live Captions is initiated (for a lack of a better term) from the Accessibility settings, but on Graphene OS there are no entries for Live Captions. I have tried to launch the app using A Shell You with a Shizuku terminal session but have not had success. Mind you I am not an Android developer, so I was turning to AI for ideas on what commands to try. On stock Android there is an option to create what is possibly a symlink or shortcut in the App Drawer. Using an APK extractor app I realized after about 10 minutes that it is not a conventional APK.
I suspect Google intentionally made Live Caption a system module due to privacy concerns since it does have access to system audio that I'm sure a lot of malevolent actors would love to tap into. The irony, considering how much other telemetry and fingerprinting data Google freely allows any app to access or request access to.
Reliable, accurate, offline real-time transcription of both system audio and microphone audio is my absolute top priority. For me, this is a non-negotiable. Let me know if there's something I can do on my end to assist with progress towards a solution. It would be great to see GrapheneOS fully inclusive for deaf and hard of hearing people, I would really like to stick with GrapheneOS if I can.