Hallelujah, I've found a FOSS transcription app that allows offline, on-device transcription of pre-recorded audio files, e. g. downloaded Signal voice messages.
Fair warning: It uses OpenAI's machine learning model "Whisper", which runs locally, but was most likely not trained ethically by OpenAI. But least it's being put to good use as a valuable accessibility tool. 🙏
"Notely Voice" by Tosin Onikute is available for Android and iOS and even on F-Droid:


👾