Transcribe

A self-contained macOS app that transcribes audio and video files locally on your Mac. No cloud upload, nothing to install. Drop a file in, get text out. Optionally paste a YouTube URL for a timestamped transcript.

Built with SwiftUI and whisper.cpp (CoreML + Metal accelerated), bundled inside the .app.

Features

• 100% offline transcription (audio never leaves the machine)
• Bundles whisper.cpp with CoreML + Metal acceleration
• Converts audio via AVFoundation (no ffmpeg required)
• Downloads the Whisper model automatically on first launch (~650MB, cached)
• Supports m4a, mp3, wav, aiff, caf, flac, mp4, mov, m4v, aac, 3gp, amr, au, ogg, oga, opus, webm
• Optionally paste a YouTube (or other video) URL to transcribe it with timestamps — the required tools are fetched on demand, not bundled

Transcribing from a URL (secondary feature)

Click "Transcribe from a URL…" under the drop zone, paste a YouTube (or other video) link, and the app downloads just the audio and produces a timestamped transcript ([m:ss] … lines). This is the one feature that reaches the network — it fetches the video you asked for; the transcription itself still runs locally.

To keep the app small (~5 MB), the helper tools this needs are downloaded on demand the first time you use the feature — not bundled:

• yt-dlp — downloads the audio stream
• deno — the JavaScript runtime recent yt-dlp needs for YouTube
• ffmpeg — decodes Opus/WebM streams when needed

They land in ~/Library/Application Support/Transcribe/tools/ (about 110 MB, one time). If a copy is already installed on the system (e.g. via Homebrew), it's used and that download is skipped. People who never touch this feature never pay for it.

Note: Downloading content from YouTube may be against its Terms of Service. Use this only for videos you own or have permission to download.

A note on Ogg / Opus / WebM

Most formats are decoded natively via AVFoundation. Ogg Vorbis, Opus and WebM can't be decoded by macOS Core Audio, so those files are transcoded with the bundled ffmpeg.

Requirements

• macOS 14.0 or later
• Apple Silicon (arm64)
• Xcode command line tools (xcode-select --install) for building

Build

./build.sh

This compiles TranscribeApp.swift and assembles build/Transcribe.app (~5 MB). To install:

cp -R build/Transcribe.app /Applications/
lsregister -f /Applications/Transcribe.app

Shipping

Push to main ships a release. .github/workflows/release.yml builds on an Apple-Silicon runner, packages a DMG, and publishes a GitHub Release. Every release includes a stable-named Transcribe.dmg, so these three channels always point at the newest build with no extra work:

• GitHub — https://github.com/J3012B/Transcribe_Mac/releases/latest
• Website — the page in docs/ (GitHub Pages) links to releases/latest/download/Transcribe.dmg
• Install script — install.sh:

  curl -fsSL https://raw.githubusercontent.com/J3012B/Transcribe_Mac/main/install.sh | bash

Versioning is automatic: the tag is v<marketing>.<run-number> (e.g. v1.0.42), where the marketing part comes from Info.plist. Bump it when you want a new minor/major:

scripts/bump-version.sh 1.1      # edits Info.plist -> next builds are 1.1.<run>
git commit -am "Release 1.1" && git push

Pull requests and non-main branches get a build check via .github/workflows/ci.yml (no release).

Signing & notarization (for a warning-free download)

Without an Apple Developer ID the DMG is ad-hoc — the website download shows a Gatekeeper prompt (right-click ▸ Open), though the install.sh path works because it strips quarantine. For a clean public download, add these repo secrets and every main push is signed + notarized automatically:

Secret	What
MACOS_CERT_P12	base64 of your "Developer ID Application" .p12
MACOS_CERT_PASSWORD	password for that .p12
MACOS_SIGN_IDENTITY	Developer ID Application: Your Name (TEAMID)
AC_APPLE_ID / AC_PASSWORD / AC_TEAM_ID	Apple ID + app-specific password + team id (notarization)

Build a signed DMG locally the same way:

export SIGN_IDENTITY="Developer ID Application: Your Name (TEAMID)"
export NOTARY_PROFILE="transcribe"   # from `xcrun notarytool store-credentials`
./build.sh && ./scripts/make-dmg.sh

One-time setup on GitHub

Enable Pages so the website goes live: Settings ▸ Pages ▸ Source: Deploy from a branch ▸ main / docs. The site then updates on every push.

Project layout

Path	Purpose
TranscribeApp.swift	The entire app (SwiftUI, single file)
build.sh	Compiles + assembles + signs the .app bundle
Info.plist	App bundle metadata (net.heliconsolutions.transcribe)
Resources/	App icon (.icns), mic glyphs, icon master
vendor/whisper/	Bundled whisper.cpp runtime (whisper-cli + dylibs)
icon/	Python scripts that regenerate the icon and glyph assets
install.sh	curl | bash installer (downloads the latest release DMG)
docs/	GitHub Pages landing page
scripts/make-dmg.sh	Packages (and optionally signs/notarizes) the DMG
scripts/bump-version.sh	Bumps the marketing version in Info.plist
.github/workflows/ci.yml	Build check on PRs / non-main branches
.github/workflows/release.yml	Push-to-main build → DMG → GitHub Release
scripts/	CLI helpers — transcribe.sh (headless whisper wrapper), notify + AppleScript

The yt-dlp / deno / ffmpeg helpers are not in the repo or the bundle — ToolsManager in TranscribeApp.swift downloads them on demand (pinned versions) into ~/Library/Application Support/Transcribe/tools/.

CLI usage

For scripted / headless transcription without the GUI:

scripts/transcribe.sh <file-path> [language] [--copy]