Features
Everything it does, and the number behind it.
Dictation apps are easy to describe and hard to trust. So here is every part of Shout with the figure it actually runs at — pulled from the app, not from a marketing meeting.
0.5sec
Most a crash can cost you
Audio is flushed to disk twice a second, so a power cut takes half a second with it — not a session.
115MB / hr
What recording all day costs
16 kHz mono, 16-bit PCM. A full working day is about a gigabyte.
99languages
Whisper's full set
All of them auto-detected; 20 you can pin so it stops guessing.
0uploads
Audio that leaves your Mac
Every engine runs on-device. There is no cloud transcription to opt out of.
Capture
It is always recording, and it writes as it goes.
Most dictation tools start recording when you press a key. Shout is already running, holding audio in a buffer that never stops — so the sentence you said before you remembered to hit record is still there.
Flushed twice a second
Every 0.5 seconds the buffer is fsync’d to disk. Nothing waits in memory for a tidy shutdown that may never come.
15-minute chunks
Audio lands in 15-minute WAV files rather than one growing blob, so a single bad write can never take the day with it.
Recovers on its own
If the app is killed mid-write, the next launch finds the orphaned recording, repairs its header and hands it back.
Built not to fail.
Audio hits the disk as you speak. If something crashes, Shout repairs the recording on the next launch and hands it back.
Crash detected00:42:16
Capture process exited mid-recording
Buffer intact00:42:17
Last write to disk 0.3s earlier
Session recovered00:42:18
Everything up to the last write is back
Transcription
Three engines. None of them is in the cloud.
Shout ships three on-device ASR stacks and lets you pick in Settings: the fast one for English dictation, the thorough one for 99 languages, and one built for Chinese, Japanese and Korean.
Parakeet — the fast one
NVIDIA’s FastConformer, compiled to CoreML and run on the Apple Neural Engine. English only. On an Apple M5, the app’s own throughput meter reads 137× real time — a minute of speech in under half a second.
Whisper — the thorough one
OpenAI’s Whisper through WhisperKit, in nine variants from tiny up to large-v3 turbo, including 4-bit quantized builds for machines that would rather not spend the memory. 99 languages by auto-detect, 20 of them pinnable.
Qwen3 — the one for CJK
Alibaba’s Qwen3-ASR, also on the Neural Engine. The engine to pick for Chinese and its dialects, Japanese, Korean, Arabic, Hindi, Thai and Vietnamese.
A live transcript as you talk
With Parakeet, a card streams your words while you are still speaking. When you stop, the whole recording is transcribed again in one pass, so the text you keep is the clean one.
A minute of speech
60 s
Parakeet transcribing it
0.44 s
Both bars are to the same scale. 137× real time on an Apple M5, read from the app’s own meter: one machine, not a lab.
Models are downloaded once and cached on your Mac. Transcription runs in a queue, so a long recording processing in the background never blocks the next thing you say.
Output
It types where you are pointed.
The transcript does not land in a scratch window for you to copy out of. It goes into the app that had focus, in the form you asked for.
8 paste actions
Paste, paste and press return, paste and save, replace everything, prepend, copy only — picked per app, so Slack gets a send and your editor does not.
Lock a target
Pin the destination app and every transcript goes there, even when you are looking at something else while you talk.
It remembers per app
The paste behaviour you pick is saved against the app you picked it in, after the first time. You configure nothing twice.
Workflows
Your prompt, wrapped around your voice.
A workflow is a prefix, a suffix and a paste action saved as one recipe. Dictate a sentence and it arrives already inside the instruction you wrote — so the thing you are talking to gets a prompt, not a fragment.
Prefix and suffix
“Rewrite this as a commit message:” in front, your constraints behind. You speak the middle; the wrapper never changes.
Bound to an app
Attach a workflow to the app it belongs to. Your editor gets the coding recipe, Slack gets the one that sounds like a person.
A store, not a blank page
Browse community workflows by category — coding, communication, productivity, creative — install one, and edit it into your own.
Rewind
Anything you said is still there.
Because capture never stopped, the timeline goes back over everything on disk. Select a span from an hour ago and transcribe it retroactively — the meeting you did not think to record is recoverable.
Transcribe the past.
Pick any stretch of the timeline and turn it into text, even an hour later. Listen back before you do.
Drag to rewind
“…so we ship the export fix Thursday, and Alex owns the onboarding copy. Let’s check in again Monday.”
How far back
Keep a day, keep a weekend, or keep everything. Audio older than 24 hours is compressed to about 11 MB an hour, so a month of it is a few gigabytes.
Files
It transcribes what it did not record, too.
Voice memos, call recordings, podcasts, a screen recording of a meeting someone else ran. If it is audio or video, Shout will transcribe it on your Mac with the same engines it uses for your own voice.
From the menu bar or Finder
Choose Transcribe File from the menu bar, or right-click a file in Finder and open it with Shout. Audio and video both work.
Several at a time
Select a batch and they queue up one after another, behind anything you are dictating, so an import never slows down what you are saying now.
Never pasted by surprise
An imported transcript goes to your history, not your clipboard and not into whatever app happens to be open. Playback plays the original file.
Meetings & calendar
It knows when the meeting starts.
Shout reads your Mac's own calendar, so anything macOS syncs in — Google, Outlook, Exchange — shows up without a single integration or OAuth screen. It reads them locally; no calendar data leaves the machine.
A nudge before it starts
A reminder 1, 5 or 10 minutes ahead of a calendar event, so the recording is running before the first person says hello.
Meeting-specific workflows
Point meetings at their own workflow — notes, actions, a summary in your format — or send the raw transcript through untouched.
Pause when you need to
Thirty minutes, an hour, two, or until you say otherwise. It is always recording because you decided it should be, not because you cannot stop it.
Echo cancelled properly
A 1024-tap adaptive filter models roughly 64 ms of the speaker-to-mic path — enough for a laptop’s geometry — and freezes adaptation while you are talking over someone, so it never erodes your own voice.
Never makes it worse
On headphones there is no echo to cancel and the canceller detects that and stands down. If cancellation cannot beat the raw microphone, the raw microphone is what you get.
Export
The recording is yours. Take it with you.
Export any recording as M4A audio at the quality you choose, and your whole transcript history as JSON. Every transcript also lands on your clipboard or straight in the app you were using.
Details
The small things you would otherwise have to remember.
Pauses your music
Spotify and Music are paused while you talk and resumed afterwards, so the recording is your voice rather than your voice over a chorus.
Global shortcuts
One key to start and stop from inside any app, plus six more for the timeline, history, copying the last transcript and locking a target. Every binding is yours to change.
Cards that get out of the way
Transcripts arrive as cards that expand when they are ready and dismiss themselves on a timer you set — or never, if you would rather they waited.
Custom vocabulary
Teach it the names, products and jargon it keeps mishearing, once.
Searchable history
Every transcript stays on your Mac and stays findable, long after the card is gone.
You can see it working
A floating strip that moves with your voice, so “is it recording?” is never a question you have to answer twice.
Stealth mode
One shortcut hides the floating bar and the transcript cards while you share your screen. Recording, shortcuts and pasting carry on, and a warning that needs you still shows.
Favourites and collections
Star the conversations that mattered and group them into collections. One conversation can live in several.
Health in the menu bar
The menu bar icon turns amber when something is degraded and red when recording has stopped, so you know even with everything else hidden.
Specification
The short version.
One price, all of it.
There is no feature tier. Everything on this page is in the app you download.