ATTESTRUM // BOOTH

A macOS recording booth for reading scripts aloud — teleprompter-style, one passage at a time, with a take stack, crash-safe lossless recording, and a clean WAV/MP3 export at the end. Built for narrating long-form video scripts; useful for anyone who records voice-over against a written script.

Everything is local. No accounts, no cloud, no telemetry. Your script, your takes, and your exports never leave your machine.

Status: pre-release. macOS 13+ on Apple Silicon (notarized .dmg) and Windows 10/11 x64 (NSIS installer; the first build is unsigned).

What it does

1. Load a script and Booth splits it into recordable passages (sentence detection, ~600-character groupings, chapter-aware). Review and adjust the grouping in the transcript screen — merge, split, pick where to start.
2. Record passage by passage. Big teleprompter text, R to record and stop, takes stack up newest-first. With more than one take, click a take to pick the kept one. Bad take? Re-record, revert (D D), or delete any take with its ✕ — files are moved, never deleted, and a 5-second undo backs every removal.
3. Trim it. Selecting a take shows its waveform — click to set the play cursor, drag to select a span, Del to cut leading/trailing/mid dead air. Cuts are non-destructive (the WAV is never touched) and applied on play and at export.
4. Edit the script inline. Click the teleprompter text, fix the wording, save — the edit writes back to your script file.
5. Export a sample-perfect concatenated voice.wav (and voice.mp3 when ffmpeg is installed), with prior outputs backed up first.

Recording is crash-safe: takes are 24-bit WAVs flushed continuously; a crash mid-take is auto-repaired on next launch. Sessions resume exactly where you left off.

Transcribe (URL or file)

The Load screen also transcribes — paste a video link (YouTube / TikTok / Instagram / Facebook) or pick a local audio/video file (⌁ FILE). It's a local, personal feature: everything runs on your machine, any downloaded media is transient and deleted immediately, and the output is transcript text only.

• URLs prefer existing captions. If the video already has a caption track, Booth imports it (no transcription compute) — instant and tagged manual-subs or auto-subs. Only when there are no captions does it download the audio and run Whisper locally (large-v3-turbo, Metal GPU; the ~1.6 GB model downloads once on first use). Caption-skip is reliable mainly on YouTube; TikTok/IG/FB usually take the Whisper path.
• Local files always transcribe with Whisper.
• Export the result as TXT, SRT, VTT, JSON, CSV, HTML, DOCX, or PDF.
• Saved transcripts appear in a TRANSCRIPTS group on the Load screen, re-openable and exportable any time.

YouTube now gates requests behind a bot check, so Booth passes your Chrome browser cookies to yt-dlp (--cookies-from-browser). Keep Chrome installed and signed in; macOS may ask once for keychain access.

Key vocabulary

Key	Action
SPACE	play / pause the selected take
R	record / stop
D D	revert newest take (double-tap; moves to discarded/)
click waveform · drag · Del	set play cursor · select a span · cut it
✕ (take card)	delete that take (single click; 5 s undo)
click a take card	select it as the kept take
U	undo a revert/delete
ENTER	accept take ▸ next passage
J / K	next / previous passage
G	view transcript (merge M / split S)
TAB	review screen
⌘E / ⇧⌘E	export / export partial
ESC	back

Every key is also a visible button — the KEY BINDINGS button (or ?) shows this table in-app. Click the device name in the top rail to switch recording input.

Install

macOS — download the latest .dmg from Releases, drag Booth.app to Applications, and launch. macOS will ask for microphone access on your first recording.

Windows — download Booth-Windows-x64-setup.exe from Releases and run it. The first build is unsigned, so SmartScreen shows an "Unknown publisher" notice — click More info → Run anyway. The installer pulls the WebView2 runtime if it isn't already present.

voice.mp3 export and mixed-sample-rate sessions use ffmpeg if it's installed (brew install ffmpeg); WAV export is built in and always works.

Build from source

git clone https://github.com/Attestrum/booth
cd booth
npm install
bash src-tauri/scripts/fetch-yt-dlp.sh   # one-time: fetch the yt-dlp sidecar
npm run tauri dev      # live dev (Rust + Vite HMR)
npm run tauri build    # release .app → src-tauri/target/release/bundle/macos/

Requires Rust (stable, via rust-toolchain.toml), Node 22+, and cmake (brew install cmake) — the transcription engine builds whisper.cpp from source with the Metal backend. The yt-dlp sidecar is git-ignored; the fetch script above pulls the latest build (re-run it periodically as sites change). The Whisper model is not bundled — it downloads to app-data on first use. To produce a signed local build, export APPLE_SIGNING_IDENTITY with a certificate from your own keychain — unsigned builds work but macOS resets the microphone permission on every rebuild.

• Audio is captured in Rust (cpal + hound, 24-bit), never in the webview.
• Sessions live in a booth/ folder next to your script: session.json (atomic writes), takes/, discarded/, replaced/ (export backups).
• Mixed input sample rates within a session are allowed; export resamples minority-rate takes to the highest take rate before concatenating.
• DESIGN.md is the spec — every screen, state, and transition is diagrammed, and changes go diagram-first.

License

Apache-2.0 OR MIT, at your option. Copyright © Hyper Beam Media LLC.

Booth is a tool from the Attestrum studio.