Transcribe a file: drop audio/video, get transcript + SRT/VTT export

[ElbertePlinio]

Turn PickScribe into a transcript source for existing media: drop or pick an audio/video file, get a transcript using the whisper.cpp engine that's already shipped — no new STT dependency.

Why

• The engine, model management, and history storage all exist; only the input path is new.
• Common ask for dictation apps, and it makes PickScribe the entry point for video-editing workflows (captions/subtitles from recorded footage).

Scope

• File input: drag-and-drop onto the app + file picker (audio and video formats)
• FFmpeg pre-step: extract/convert audio to 16 kHz mono WAV (whisper.cpp's expected input); clear error if ffmpeg is missing
• Transcribe via existing stt.rs path (transcribe_with_cancel), with progress + cancel in UI (files are much longer than dictations)
• Result lands in History like a normal session, marked as file-sourced (original filename)
• Export: plain text, SRT, VTT (whisper.cpp --output-srt / --output-vtt); word-level timestamps where the engine supports it
• Local-only mode respected: file transcription is fully local; cleanup step optional and off by default for files

Out of scope

• Speaker diarization
• Editing/cutting media
• Batch/folder transcription (follow-up if wanted)

Validation

• Transcribe a short WAV, an MP3, and an MP4 (audio extracted); verify History entry + all three export formats
• Kill/cancel mid-transcription leaves no orphan temp files
• Local-only mode: no network calls during file transcription
• cargo test / existing suite stays green

Current status: Planned
Next action: not scheduled — created from the ai-video-editing research session (agents need transcripts of recorded footage; PickScribe already has the engine)

[ElbertePlinio]

Plan (progress tracker)

Split into 2 PRs so each reviews clean:

PR 1 — Rust core (no user-visible change) · feat/file-transcription-core

• engine/media.rs: ffmpeg → 16 kHz mono WAV, cancellable, clear error if ffmpeg missing
• engine/stt.rs: file transcription returning timestamped segments (--output-json) + progress callback
• Export formatters: TXT / SRT / VTT from segments (unit-tested; single whisper pass, export anytime from History)
• history.rs: additive migration — source_file, segments_json
• cargo test --workspace green

PR 2 — App glue + UI · feat/file-transcription-ui (after PR 1 merges)

• Commands: transcribe_file, cancel, export; progress events
• Drag-and-drop + file picker (dialog plugin)
• Progress + cancel UI; temp files cleaned on cancel/kill
• History entry marked file-sourced (original filename) + TXT/SRT/VTT export via save dialog
• Cleanup step off by default for files; local-only fully respected
• bun run check / test / cargo test green, UNRELEASED.md updated

Notes: segment-level timestamps in v1 (word-level = follow-up if wanted); SRT/VTT generated by our own tested formatters from whisper JSON instead of whisper's --output-srt/--output-vtt files — one whisper pass, re-export anytime.

Next action: PR 1 in progress.

[ElbertePlinio]

Progress update:

• Confirmed PR 1 scope: Rust core only
• Add media conversion, transcript formats, file STT progress, and history migration
• Run workspace Rust tests and formatting

PR: pending
Blocked: none
Next: implement the core slice

[ElbertePlinio]

PR 1 open: #32 (core: ffmpeg convert, whisper segments + progress, TXT/SRT/VTT formatters, history migration). Panel-reviewed, cargo suite green (74 tests). Codex review requested. PR 2 (commands + UI) starts on top of it now.

[ElbertePlinio]

Agent plan

Goal: implement the Rust Tauri command/event glue for file transcription on this branch.

Checklist:

• Add Rust dialog plugin and capability
• Add cancellable file-job worker/events/history persistence
• Add picker, transcribe, cancel, and export commands
• Add ffmpeg doctor check and release-note update if needed
• Run workspace Rust tests

Validation:

• cargo test --workspace --locked --all-targets

Current status: In progress
Next action: implement the Tauri layer

[ElbertePlinio]

PR 1 (#32) merged — Codex clean on round 1, CI green. Core checklist items done: ffmpeg pre-step, segment transcription, TXT/SRT/VTT export formatters, history migration. PR 2 (commands + UI) in progress.

[ElbertePlinio]

Progress update:

• Rust dialog plugin, capability, picker, and save/export commands
• Cancellable file-job worker with conversion/transcription/optional cleanup/history events
• ffmpeg doctor check and HistoryDb::get
• cargo check (src-tauri)
• cargo test --workspace --locked --all-targets (78 passed)
• Drag-and-drop/file-transcription UI

PR: pending
Blocked: none
Next: UI consumer can wire the documented Rust command/event contract.

[ElbertePlinio]

PR 2 open: #33 (commands + UI — drag-drop, picker, progress+cancel, History badges, TXT/SRT/VTT export, cleanup off by default for files). Panel + anti-overengineering gate passed, all suites green (cargo 79 / vitest 19 / svelte-check 0 errors). Codex review requested. Closes the issue when merged.

[ElbertePlinio]

Done — PR 2 (#33) merged, Codex clean on round 1 for both PRs. Feature complete: drop/browse audio+video → local whisper transcript with progress+cancel → History entry (file-badged) → TXT/SRT/VTT export. Cleanup off by default for files, metrics unaffected by file jobs, ffmpeg doctor check added.

Deferred (noted in #33): cancel-during-cleanup latency, multi-file drop takes first only, word-level timestamps. Interactive drag-drop smoke in a live desktop session is the one remaining manual check.