Does WhisprDesk transcription work offline?

Yes, once your chosen model has been downloaded. Transcription runs on-device with Whisper or NVIDIA Parakeet, so no connection is needed to process audio. A licensed copy does re-check its licence periodically, so a permanently offline machine is not the intended setup, but ordinary offline use works.

How much disk space do the models need?

It depends which you pick. Whisper Tiny is about 75MB and Base about 142MB, through to Large at about 3GB. NVIDIA Parakeet is about 680MB. The installer itself contains no model weights; they download on first use and cache under ~/.cache/whisprdesk/.

Which formats are not supported?

.mov, .mkv, .avi and .m4v will not open. Extract the audio track first and WhisprDesk will transcribe the result. Supported locally: mp3, wav, m4a, webm, ogg, flac, aac, mp4, wma and opus.

Can it transcribe languages other than English?

Yes. NVIDIA Parakeet TDT 0.6B covers 25 languages and is the one to try first for multilingual audio. Whisper is also multilingual, with the larger sizes generally handling non-English audio better than the smaller ones.

Do I need an account to transcribe?

No. WhisprDesk has no accounts and no sign-in. Transcription runs locally on your Mac.

Is there a per-minute cost to transcribe?

No. Because transcription happens on your own machine, there is no cloud usage meter. WhisprDesk is a one-time purchase with no subscription; only the optional AI cleanup uses your own API key, billed by your provider.

What are the system requirements?

WhisprDesk runs on Apple Silicon Macs (M1 or later) on macOS 12 (Monterey) and later.

Transcribe audio files on your Mac, offline

WhisprDesk transcribes audio files on your Mac without sending them anywhere. You open an audio file, and the transcription runs on-device using Whisper or NVIDIA Parakeet, so the audio never leaves your machine: no upload, no cloud queue, no account. That makes it a fit for recordings you would not want on someone else's server, like interviews, meetings, and personal notes. Because the work happens locally on Apple Silicon, there is no per-minute cloud bill and no file sent over the network. If you want cleaned-up text afterward, an optional step can tidy the transcript using your own OpenAI, Anthropic, or Google API key, so even that runs on your terms. WhisprDesk is available now for macOS 12 and later, with a 7-day free trial.

How on-device transcription works

WhisprDesk runs the speech-to-text model on your own Mac. You pick one of two engines, Whisper or NVIDIA Parakeet, and the audio is processed locally rather than uploaded to a server. Nothing about the recording is sent over the network to be transcribed.

Running locally means transcription works without an internet connection once your chosen model is downloaded, and there is no usage meter counting the minutes you process. A licensed copy does check its licence periodically, so it is not built for a permanently disconnected machine, but day to day, on a plane or on a train, it simply works.

What file types can I transcribe?

Transcribing locally, WhisprDesk accepts mp3, wav, m4a, webm, ogg, flac, aac, mp4, wma, and opus. You open the file and the transcription runs on your machine.

Some container formats are not supported: .mov, .mkv, .avi, and .m4v will not open. If you have a recording in one of those, extract or convert the audio track first and WhisprDesk will handle the result.

If you switch the app to use a cloud model with your own API key, the accepted list is narrower (mp3, wav, m4a, webm, ogg, flac, aac) and files are capped at 25MB. Local transcription has no file-size cap beyond your own disk.

Which model should you choose?

WhisprDesk ships support for six sizes of OpenAI's Whisper model plus NVIDIA's Parakeet. Larger models are more capable and slower, and they use more disk and memory. Base is the default starting point for most people.

Whisper Tiny is about 75MB, Base about 142MB, Small about 466MB, Turbo (large-v3-turbo) about 1.6GB, Medium about 1.5GB, and Large (large-v3) about 3GB. NVIDIA Parakeet TDT 0.6B is about 680MB and covers 25 languages, which makes it the one to try first for multilingual work.

You can hold more than one and switch between them. A smaller model for quick notes and a larger one for an interview you care about is a reasonable way to work.

What happens the first time you run it

The download you install contains the application, not the speech models. Model weights are fetched the first time you select a model, and cached under ~/.cache/whisprdesk/ on your Mac.

That means the installer stays small, you only store the models you actually use, and you need a network connection once per model. After that the model is on your disk and transcription no longer needs one.

Is my audio uploaded anywhere?

No. Transcription is on-device, so the audio file is never uploaded for processing. There is no account and no telemetry. The only time anything leaves your Mac is if you choose the optional AI cleanup step, which sends text (not your audio) directly to the AI provider whose key you supply.

Optional AI cleanup with your own key

After a transcript is produced locally, you can optionally have an AI model tidy it up, fix punctuation, or reformat it. That step uses your own OpenAI, Anthropic, or Google API key, so the request goes from your Mac directly to the provider on your account. It is entirely optional; transcription itself needs no key.

Frequently asked questions

Does WhisprDesk transcription work offline?: Yes, once your chosen model has been downloaded. Transcription runs on-device with Whisper or NVIDIA Parakeet, so no connection is needed to process audio. A licensed copy does re-check its licence periodically, so a permanently offline machine is not the intended setup, but ordinary offline use works.
How much disk space do the models need?: It depends which you pick. Whisper Tiny is about 75MB and Base about 142MB, through to Large at about 3GB. NVIDIA Parakeet is about 680MB. The installer itself contains no model weights; they download on first use and cache under ~/.cache/whisprdesk/.
Which formats are not supported?: .mov, .mkv, .avi and .m4v will not open. Extract the audio track first and WhisprDesk will transcribe the result. Supported locally: mp3, wav, m4a, webm, ogg, flac, aac, mp4, wma and opus.
Can it transcribe languages other than English?: Yes. NVIDIA Parakeet TDT 0.6B covers 25 languages and is the one to try first for multilingual audio. Whisper is also multilingual, with the larger sizes generally handling non-English audio better than the smaller ones.
Do I need an account to transcribe?: No. WhisprDesk has no accounts and no sign-in. Transcription runs locally on your Mac.
Is there a per-minute cost to transcribe?: No. Because transcription happens on your own machine, there is no cloud usage meter. WhisprDesk is a one-time purchase with no subscription; only the optional AI cleanup uses your own API key, billed by your provider.
What are the system requirements?: WhisprDesk runs on Apple Silicon Macs (M1 or later) on macOS 12 (Monterey) and later.