Tom: Transcribe Audio & Video

Engine choice — Tom now ships three transcription engines, all on-device. Pick the one that fits the file: • Apple Speech — fast, built into iOS, great default. • Whisper — six sizes (Tiny, Base, Small, Medium, Turbo, Large v3) up to 99 languages. • Parakeet v3 — high-accuracy European transcription via FluidAudio. Re-transcribe with a different engine — not happy with the result? Re-run the same audio through another engine in one tap. The previous transcript is kept in a per-file version history so you can compare and switch back at any time. Settings, redesigned — System / Advanced sections, language filter that greys out engines that don't speak your locale, inline download button with real progress and clear error messages. Honest hallucination flag — when Whisper invents text on silence or music, suspect segments are now clearly marked in the detail view. We flag, we never silently strip. Model management from Files — downloaded Whisper and Parakeet models live in the Documents folder. Browse, audit, and delete them straight from the Files app, or use the new "Downloaded models" list in Settings. Wi-Fi only downloads — optional toggle (on by default) so large model downloads never run on cellular. Quality and reliability — robust Whisper download with disk-space pre-checks, automatic ANE→GPU fallback on older devices, special-token cleanup, distinct "Loading model…" phase indicator, and a Cancel button that actually stops mid-run. Requires iOS 17. Built for iPhone and iPad.