v2.4.0

Changes

• Transcribing no longer locks you out of the app. A running job used to replace the whole window, so for the several minutes an hour-long meeting takes, you could not read the last transcript, look a quote up, or export anything. Progress now sits in a band across the top and the rest of the app stays where it was and stays usable.
• A transcript that finishes while you are reading something else no longer yanks you out of it. The bar offers it and you open it when you want. If you were on the home screen waiting for it, it still opens by itself.
• Cancelling a job, or a job failing, no longer throws you back to the home screen from whatever you had open.

Both platforms.

Downloads

Windows 10 1803 or later, 64-bit

• VerbatimSetup-2.4.0.exe — installer
• Verbatim.exe — portable, single file, no install

macOS 14.4 (Sonoma) or later, Intel or Apple silicon

• Verbatim-2.4.0.dmg — open and drag Verbatim to Applications
• Verbatim-2.4.0.zip — the same app, unzipped

Nothing is notarised. Windows SmartScreen may ask once: "More info" then "Run anyway". macOS Gatekeeper may ask once: right-click the app, then "Open", then "Open".

Full changelog: CHANGELOG.md · v2.3.1...v2.4.0

v2.3.1

Fixes to the macOS app from 2.3.0. Windows is unchanged apart from the version number.

Changes

• The macOS app has its icon. 2.3.0 shipped with the generic blank one: the build script copied an icon file that was never actually created.
• Renaming a speaker is findable now. The bar above the transcript says "Click a name to change it", each chip shows a pencil when you point at it, and clicking turns the chip into the name field in place. Previously the editor opened in a box anchored to the wrong part of the window.
• Right-click a line to rename whoever said it, mark it as a different speaker, copy just that line, or jump the audio there.

Downloads

Windows 10 1803 or later, 64-bit

• VerbatimSetup-2.3.1.exe — installer
• Verbatim.exe — portable, single file, no install

macOS 14.4 (Sonoma) or later, Intel or Apple silicon

• Verbatim-2.3.1.dmg — open and drag Verbatim to Applications
• Verbatim-2.3.1.zip — the same app, unzipped

Nothing is notarised. Windows SmartScreen may ask once: "More info" then "Run anyway". macOS Gatekeeper may ask once: right-click the app, then "Open", then "Open".

The first time you record on macOS it asks permission to use the microphone and to record the audio your Mac is playing. Both are needed to capture both sides of a call.

Full changelog: CHANGELOG.md · v2.3.0...v2.3.1

v2.3.0

Changes

• The transcript appears while you record. The record window now shows the meeting as it happens, a few seconds behind whoever is speaking, instead of only after you stop. Turn it off with a checkbox if you would rather the machine just recorded. It is a preview: no speaker names, and each piece is recognised without the words around it. When you stop, the whole recording goes through the accurate pass as before, and that is what gets saved.
• macOS is back, as a native Swift/SwiftUI app rather than a revival of the Electron line that ended at 1.1.0. Same engine, same features, same transcript format and location, so a library copied from a PC opens untouched. Records your microphone, the audio your Mac is playing, or both. Universal build, Intel and Apple silicon.
• Markdown export (.md) — one section per speaker turn, with a timestamp on every line.
• Speech models now download before recording starts rather than after it finishes, so a first-run download cannot interrupt a meeting.
• The end-to-end tests build their own two-speaker clip instead of reading one from a scratch directory, so they run again rather than skipping silently.

Downloads

Windows 10 1803 or later, 64-bit

• VerbatimSetup-2.3.0.exe — installer
• Verbatim.exe — portable, single file, no install

macOS 14.4 (Sonoma) or later, Intel or Apple silicon

• Verbatim-2.3.0.dmg — open and drag Verbatim to Applications
• Verbatim-2.3.0.zip — the same app, unzipped

Nothing is notarised (I am still not paying Apple 99 dollars a year). Windows SmartScreen may ask once: "More info" then "Run anyway". macOS Gatekeeper may ask once: right-click the app, then "Open", then "Open".

The first time you record on macOS it asks permission to use the microphone and to record the audio your Mac is playing. Both are needed to capture both sides of a call. They live in System Settings, Privacy and Security.

Full changelog: CHANGELOG.md · v2.2.0...v2.3.0

v2.2.0

Full Changelog: v2.1.0...v2.2.0

v2.1.0

Five additions, all about living with real transcripts.

New

• Fix the diarizer's mistakes — right-click any line to reassign it to another speaker (or a brand-new one), copy it, or delete it. Right-click a speaker chip to merge that speaker into another — perfect for when two "speakers" are really one person.
• Search your whole library — the home screen now searches titles, speaker names, and every word of every transcript you've made, with a snippet of the matching line.
• Word export — Export → Word (.docx): timestamps, colored speaker names, ready to share.
• Playback speed is back (0.75×–2×) — and unlike 1.x it keeps the pitch natural instead of chipmunking.
• Arrow-key seeking — ←/→ skip 5 seconds; Space still play/pauses; Ctrl+F still searches.

Windows 10 1803+ (64-bit) · unsigned as always (still not paying for a certificate) · upgrading in place keeps your library, settings, and models.

v2.0.0

Verbatim is now a native Windows app — a full rewrite in C#/.NET 9 WinForms, the same stack as Photon. No more Electron.

Downloads

File	What it is
VerbatimSetup-2.0.0.exe	Installer (per-user, no admin needed)
Verbatim.exe	Portable — single file, no install, engines included

Windows 10 1803+ (64-bit). Unsigned (I am NOT paying for a code-signing certificate) — if SmartScreen asks: "More info" → "Run anyway".

What you get

Everything from 1.x, now native: automatic speaker detection with renaming, click-to-jump playback with waveform and live transcript follow, search with in-text highlights (Ctrl+F), double-click editing, TXT/SRT/VTT/JSON export, an auto-saved library, and transcribe-from-a-link (YouTube/podcasts/any URL via yt-dlp with one-click updating in Settings).

Your data carries over. Same storage location and format as 1.x — existing transcripts, settings, and downloaded models are picked up as-is.

Notes

• Audio formats: MP3, WAV, M4A, AAC, WMA, FLAC (native Windows decoding). URL fetches automatically pick a compatible format.
• Transcription stays 100% local: Whisper + pyannote via bundled sherpa-onnx engines; models download once (~250 MB) on first use.
• macOS: the Electron line ended with v1.1.0, which remains available.
• Known gap vs 1.x: no playback-speed selector yet.

v1.1.0

New: transcribe straight from a link. Paste a YouTube URL, a podcast episode, or any direct audio/video link on the home screen — Verbatim fetches the audio and transcribes it like any local file, speakers and all.

Downloads

Platform	File	Notes
macOS	Verbatim-1.1.0-mac-x64.dmg	Intel + Apple Silicon. Unsigned — right-click → Open on first launch.
Windows	Verbatim-1.1.0-win-x64.exe	Installer, Windows 10 1803+
Windows	Verbatim-1.1.0-win-x64.zip	Portable

What's new in 1.1.0

• Transcribe from a URL — YouTube, podcasts, and thousands of other sites via yt-dlp (downloaded automatically on first use), plus any direct audio/video link. The video/episode title becomes the transcript title.
• Fetched audio is stored with your library so playback and click-to-jump keep working later, and is cleaned up when you delete the transcript.
• Settings → Link downloader shows the yt-dlp version with a one-click Update button — if links ever stop working, update and retry.
• Whisper's non-speech markers ([Music], (applause), ♪) are filtered out of transcripts.
• Guardrails: live streams are refused; links longer than 4 hours or over ~1.4 GB are refused before download; playlist links take the first entry only.

Full details in CHANGELOG.md.

Transcription remains 100% local — the network is used only for links you paste and one-time model downloads.

v1.0.0

Verbatim — every word, on your machine. Free, private, offline audio transcription with automatic speaker identification, for Windows and macOS.

Downloads

Platform	File	Notes
macOS	Verbatim-1.0.0-mac-x64.dmg	Intel + Apple Silicon (via Rosetta shell; transcription engines run native on both). Unsigned — right-click → Open on first launch.
Windows	Verbatim-1.0.0-win-x64.exe	Installer, Windows 10 1803+
Windows	Verbatim-1.0.0-win-x64.zip	Portable, no install needed

Features

• Transcription that never leaves your computer — Whisper speech recognition running locally via sherpa-onnx. No account, no cloud, no fees.
• Speaker identification — pyannote diarization figures out who spoke when. Click a speaker to give them a real name; it applies everywhere and is saved.
• Click to jump — click any line's timestamp to hear that exact moment. The transcript highlights and follows the audio as it plays, with waveform seek bar and playback speed control.
• Search — Cmd/Ctrl+F, match counter, Enter to hop between matches; each jump seeks the audio too.
• Edit and organize — double-click a line to fix a word; transcripts auto-save to a library on the home screen.
• Export — TXT, SRT, VTT, JSON, clipboard, or a portable project file you can open on another machine.
• Formats — MP3, WAV, M4A, AAC, FLAC, OGG, OPUS, WEBM.
• Models & languages — Whisper Tiny / Base / Small (pick in Settings), ~30 languages with auto-detect, or fix the number of speakers. Models download once on first use (~250 MB), then everything is offline.

Notes

• First transcription downloads the speech models (one time, with progress shown).
• Builds are unsigned (free software, no certificates): macOS Gatekeeper needs right-click → Open once; Windows SmartScreen needs "More info" → "Run anyway".