GrapheneOS on Nostr: nprofile1q…3xc50 We aren't sure how those ones were trained. We want to stick with ...

nprofile1qy2hwumn8ghj7un9d3shjtnddaehgu3wwp6kyqpqck0t5glr98tjvvlvlxv950g9l8j0kqq3hma70sd3dss228z7ffnsd3xc50 (nprofile…xc50) We aren't sure how those ones were trained. We want to stick with ones trained using open data for what we bundle within the OS, but people could switch to one of the better models without clear provenance for the training data if they wanted. We likely just want to include a fork of the app modified to have the model we decide on using bundled and set up by default so there's working text-to-speech for TalkBack to use from the 1st page of the setup. It just needs to be tolerable.