FROM: your recording
TO: your Mac
VIA: the Neural Engine

Your recordings stay on your Mac.

Transcripts with speaker labels and summaries, right on your Mac. Even offline.

Mac with Apple silicon, macOS 15 or later

One hour of audio in about 30 seconds (measured on an M5 Mac)

Where everything goes
What	Where
The audio	~~not an online service~~ your Mac
The transcript	~~not a cloud server~~ your Mac
Who's speaking	your Mac
The summary	your Mac, with Apple Intelligence

Or OpenAI with your own key, only if you choose it: the text is sent, not the audio.

Drop in the file. Your Mac does the rest.

Works with files exported from Plaud and other recorders, with voice memos, and with any audio file in a supported format.

Drag the file into the app

MP3, M4A, WAV, FLAC, Ogg/Opus, or the audio of a video.
Your Mac transcribes it and tells the speakers apart

With the Parakeet model, on your Mac's Neural Engine. Your audio is never uploaded.
Review, summarize, export

Read it through, rename the speakers, ask for a summary and export it.

Summaries use Apple Intelligence on your Mac (macOS 26) or, if you choose, OpenAI with your own key: in that case the text, not the audio, is sent to OpenAI.

Export to:

Text
Markdown
SRT
VTT
JSON
Clipboard

Sample transcript meeting.m4a

Speaker 1 00:00:04

Let's start with the schedule: is delivery still set for the end of the month?

Speaker 2 00:00:09

Yes, as long as we close the revisions on chapter two by Friday.

Speaker 1 00:00:15

Then I'll mark it as a priority. Who's taking it?

Speaker 3 00:00:19

Me. I'll send you the draft on Thursday evening.

Sample text written for this page, not taken from a real recording.

What stays on your Mac. And what doesn't.

The transcription engine runs on your computer. Below is everything the app sends over the network.

Stays on your Mac

The audioon your Mac
The transcripton your Mac
Who's speakingon your Mac
Summaries with Apple Intelligenceon your Mac

Goes online, only when needed

The transcription models, about 1 GBfrom Hugging Face, once
Update checkssite download only
Summaries with OpenAIonly if you choose it, with your key: the text is sent, not the audio

No account. No usage analytics, no tracking.

Read the app privacy policy

Fast, even on long recordings.

Times measured on an M5 Mac, from audio file to transcript with speaker labels. Times vary on other Macs.

Measured on one M5 Mac. Transcription and telling speakers apart run in parallel.
Audio	Transcription	Mac
33min	17.5s	M5
60min	about30s	M5

Mac: With Apple silicon
System: macOS 15 or later
On-device summaries: macOS 26 with Apple Intelligence turned on
Storage: About 1 GB for the models, downloaded once, plus space for your transcripts
Languages: 25 European languages
Files: MP3, M4A/AAC, WAV, AIFF, CAF, FLAC, Ogg/Opus (macOS 26 and later), audio from MP4 and MOV
Network: Only to download the models and check for updates; transcription works offline

Questions

Does it work on my Mac?

You need a Mac with Apple silicon and macOS 15 or later. On-device summaries need macOS 26 with Apple Intelligence turned on; without it, you can use OpenAI with your own key, or transcription only.

Does it work offline?

Yes, after the first launch. The first time, the app downloads the transcription models, about 1 GB, once. From then on, transcription and summaries with Apple Intelligence also work offline.

Do I need OpenAI?

No, it's optional. Summaries use Apple Intelligence on your Mac. If you prefer OpenAI, you choose it and use your own key: in that case the transcript text, not the audio, is sent to OpenAI under the terms of your account, and you pay for the usage.

How accurate is it?

It depends on the audio quality and the language. Review the transcript before you rely on it.

How good are the Apple Intelligence summaries?

Apple Intelligence works on your Mac with a small context: it suits short and medium conversations, while in long meetings it can miss details. Review the summary before you use it.

Which languages does it recognize?

25 European languages, including English and Italian, and it detects the language on its own. Languages such as Chinese, Japanese and Arabic aren't supported.

Can Waxrose import recordings directly from my Plaud account?

Waxrose works with audio files on your Mac. To transcribe a Plaud recording, export it as MP3 with Plaud's Export feature and drag the file into Waxrose. If you have many recordings, Plaud's official command-line tool (Plaud CLI) can download your recordings into a folder for you; then drop the files into Waxrose. Waxrose is not affiliated with Plaud.

What counts as a transcription?

Each file you transcribe. Reviewing and exporting transcripts you've already made is always free, in the free version too.

If I buy on the Mac App Store, does it also cover the download from this site?

No. The Mac App Store and this site have separate purchases: the unlock applies to the app from the channel where you bought it.

Do I need an account?

No. You don't need to sign up for or sign in to anything to use Waxrose.

Pricing

Use it every day without paying. If you need more, remove the limit once.

Free

€0

One transcription a day, with speaker labels, summary and export.

Download for Mac

Full

€14.99 one-time

Launch price €14.99 until [date], then €19.99. VAT included where applicable.

Unlimited transcriptions, one-time purchase, no subscription.

Unlock: €14.99 until [date] (then €19.99)

On this site, payment is handled by Paddle. On the Mac App Store, the unlock is an in-app purchase: the two channels have separate purchases.

Download Waxrose

Your recordings, sealed on your Mac.

Download for Mac

Signed with Developer ID and notarized by Apple
Mac with Apple silicon, macOS 15 or later
About 1 GB of models on first launch, once
On-device summaries with macOS 26 and Apple Intelligence
One transcription a day for free, then a one-time unlock
No account, no subscription

Credits and licenses

Parakeet Ultra by moondream, derived from NVIDIA's Parakeet TDT 0.6B v3CC BY 4.0
Speaker diarization models (pyannote, WeSpeaker, BUT Speech@FIT), converted by Fluid InferenceCC BY 4.0
FluidAudio by Fluid InferenceApache 2.0
Archivo and Courier Prime typefacesSIL OFL 1.1

Full acknowledgements