Features

Everything Transcribbler does.

A native Mac app that turns anything you can hear into searchable, speaker-labelled text — and does the whole job on your own machine. No account, no upload, no per-minute meter.

Transcribbler is in active development and not yet released. Everything on this page is built and working in the current build. Things that are planned but not finished are not listed here — if you can't find a feature on this page, assume it doesn't exist yet.

Getting audio in

Ten ways in, however the sound reaches you.

Files you already have, a live microphone, whatever's playing through your speakers, or a folder that quietly transcribes itself.

Drop in files

Audio or video — mp3, wav, m4a, mp4, mov. Drag a whole batch in and transcribe them together.

Record your microphone

Press record and capture straight from your mic. Pause and resume mid-recording, and drop a bookmark at any moment you'll want to find again.

Record system audio

Capture whatever is playing on your Mac. macOS asks for Screen Recording permission once — that's the only route any app has to system audio. Transcribbler takes the sound only, never a picture of your screen.

Capture one app only

Record just the audio from a single application, so the rest of your Mac stays out of the recording.

Meeting mode

Record your mic and the call audio at the same time, mixed into one recording, then split back into you and each other participant.

Recording presets

Meeting, Lecture, Interview or Voice Note — each sets the engine, speaker detection and live transcript sensibly so you don't configure anything.

Watch a folder

Point Transcribbler at a folder. Anything new that lands there transcribes itself, hands-off.

Menu-bar quick record

A fast voice note from the menu bar that files itself into your library.

System-wide dictation

Press ⌥⌘D in any app, speak, and the text lands on your clipboard.

Live captions

A floating, always-on-top window that transcribes in real time as you speak or listen.

Automatic call detection

When a call starts in Zoom, FaceTime or Microsoft Teams, a small corner prompt offers to record it. Always dismissible, and switch-off-able entirely.

Straight from Finder

Right-click any audio file → Services → Transcribe, use Open With, or drop files onto the Dock icon. Transcribbler doesn't need to be open first.

Shortcuts & automation

Transcribe a file, start or stop recording, fetch your latest transcript or search your library — all as Shortcuts actions you can chain into your own workflows.

From a link

Paste a direct link to an audio or video file and Transcribbler fetches and transcribes it.

Accuracy

Several engines. You pick the trade-off.

Different recordings want different things. A quick voice note doesn't need the model that takes four times as long — an important interview does.

Multiple on-device engines

OpenAI Whisper from Tiny through Large v3, NVIDIA Parakeet TDT in English and multilingual flavours, and Apple's built-in speech recognition.

Fastest, or most careful

Switch engine per recording. The size and speed of each model is shown before you commit to it.

Custom dictionary

Teach it the names, medications and jargon it keeps getting wrong, so it stops getting them wrong.

It learns your corrections

Fix the same misheard word a few times and Transcribbler notices, offers to remember it, and tells you when it has. It never changes its behaviour behind your back.

Hallucination filter

Speech models invent text over silence — sign-offs like "Thank you." they picked up from subtitle training data. Transcribbler strips the ones it can identify, and skips transcribing silence in the first place.

Per-segment confidence

See which lines the engine was unsure about, so you know exactly where to double-check.

Word-level timing

Every word carries its own timestamp, which is what makes precise search, audio editing and highlighted playback possible.

Punctuation repair

Tidies up sentence breaks on engines that return unpunctuated text.

Translate to English

Transcribe audio in another language and translate it into English in the same pass.

An honest word about accuracy. No speech model is perfect, including these. They mishear words and can occasionally invent them, especially on quiet or noisy audio. Check anything you'll rely on against the recording — the confidence indicators and click-to-play exist to make that quick.

Speakers

Know who said what.

Speaker separation runs on your Mac too — no cloud service, no uploading a voice sample to a server.

Automatic speaker detection

On-device diarization labels each voice as it goes, with no setup at all.

Voice profiles

Enrol someone's voice once and they're recognised by name in every recording afterwards.

Voice training

A short guided read-aloud that teaches Transcribbler your own voice, so you're labelled correctly from the first meeting.

Rename, reassign, merge

Fix speaker labels by hand — one line or every line at once.

Talk-time breakdown

A colour bar showing who spoke for how much of the conversation, with percentages.

Working with a transcript

Built to be read, corrected and navigated.

Click any line to hear it

Jump to that moment in the audio, with a waveform scrubber for moving around by ear.

Playback, your way

Variable speed from 0.5× to 3× with pitch correction, loop a segment, and skip silence automatically.

Chapters

Long recordings are broken into sections you can jump between, detected from natural breaks in the conversation.

Find and replace

Across the whole transcript, with optional case sensitivity.

Star the lines that matter

Then filter down to only those, so a two-hour meeting becomes the six lines you needed.

Filler-word filter

Hide the ums and ahs without deleting them, and a compact mode for when you just want to read.

Rename the conversation

Click the title and type. Auto-generated names are a decent start but rarely what you'd search for later.

Notes of your own

A free-text note per conversation, for what the title can't say — what it was for, who it was with, what to do next.

Word cloud

A quick picture of what a conversation was actually about.

Merge recordings

Stitch several recordings into one continuous transcript. Non-destructive — the originals stay put.

Undo

Bulk changes like a speaker rename across every line can be taken back.

Editing the recording

Cut the audio, and the transcript follows.

These change the recording itself rather than making a copy, and the transcript is re-timed to match so timestamps stay correct.

Trim long silences

Remove the dead air from the recording and the transcript together.

Remove filler words from the audio

Not just hidden in the text — actually cut out of the recording.

Delete a line and its audio

Cut a single line out of both the transcript and the recording.

Because these rewrite the audio file, they can't be undone. Transcribbler tells you that before it does anything, every time.

Your library

Every conversation, findable later.

The part most transcription tools skip. A transcript you can't find again might as well not exist.

One dashboard

Title, participants, duration, tags and a snippet of every conversation you've had.

Filter, sort, search

By person, tag, date, length, engine or language — plus full-text search across everything.

A view per person

Click someone to see every conversation you've had with them, how long you've spent talking, and when you last spoke.

Titles and tags, written for you

Generated automatically so the library stays legible without busywork — and editable whenever they're wrong.

Optional AI

Off by default. Your own key. Never your audio.

These are the only features that use the internet at all, they're switched off until you turn them on, and they need an API key you supply. Only transcript text is ever sent — never a recording.

Summary

An overview, the key points, action items with who committed to them, and the decisions that were made.

Ask

Put a question to a conversation and get an answer with timestamp citations, so you can jump to the line and check it yourself.

Discover

Search by meaning across every conversation. Search "money worries" and it finds "can't afford it" and "bills piling up". Only your query leaves your Mac.

A log of every AI call

See exactly what was sent, when, and to whom. Trust should be checkable, not assumed.

You bring a key from Anthropic or OpenAI and pay them directly for what you use — there's no markup and nothing resold. A language model reading your transcript can still be wrong, which is exactly why answers come with citations.

Privacy & sensitive recordings

Designed for the recordings that matter most.

A doctor's appointment, a therapy session, a difficult call. The reason the whole app works locally is so these stay yours.

Redaction

Detects names, phone numbers, emails, addresses, card numbers and social security numbers. Redacting the transcript is reversible and keeps redacted terms out of exports, AI prompts and search.

Redact the recording too

Silence a person or a detail in the audio itself. This rewrites the file and can't be undone, and Transcribbler says so first.

Encrypt your library

Lock your transcripts behind a key held in your Mac's Keychain.

Lock the app

Require Touch ID or your password to open Transcribbler.

Secure erase

Overwrite a recording before deleting it, rather than just unlinking the file.

Consent reminder

Set your state and Transcribbler warns you before recording where the law requires everyone's consent. Explicitly not legal advice, and switch-off-able.

Backup and restore

Take an encrypted archive of your whole library, on a schedule or on demand, to somewhere you choose.

Everything stays local

Recordings, transcripts and voice profiles all live on your Mac — yours to move, back up or delete.

Read the full privacy policy →

Getting transcripts out

Take them anywhere.

Seven formats

Plain text, SRT and VTT subtitles, CSV, Markdown, RTF and PDF.

Bulk export

Many conversations at once, in whichever format the next tool needs.

Template editor

Design your own export format with header, body and footer sections and tokens, so output drops straight into your workflow.

Copy as rich text

Keeps speaker colours when you paste into Mail, Pages or Word.

Print

A properly laid-out printed transcript, not a screenshot of a window.

Web page with audio

One self-contained file holding the transcript and the recording. Click any timestamp to jump.

Shareable quote cards

Turn a line into an image, or a short clip into a captioned audiogram.

Send to your notes app

Hand a transcript straight to the note-taking apps you already have installed.

The app itself

A real Mac app.

Native Swift and SwiftUI

Not a web page in a window. Sandboxed, and it follows the platform conventions you already know.

Three languages

English, Spanish and French throughout the interface.

Themes

Light, dark and a handful of accent choices.

Rearrange the sidebar

Put the sections you use where you want them, and hide the ones you don't.

Crash-safe recording

Audio is written to disk continuously as you record, so a crash or a force-quit doesn't cost you the recording.

Works offline

Everything except the optional AI features works with no internet at all, including on a plane.

Requirements

macOS 14.4 or later, on an Apple Silicon Mac. Speech models are downloaded on first use and vary in size — the smallest is a few tens of megabytes, the largest around 600 MB.

Coming soon for Mac.

One email, once, the day it's available. No newsletter, no drip campaign.

Get launch news →