Is local dictation more private than cloud?

Yes. With on-device dictation your audio is processed on your own Mac and never uploaded, so there is no server-side copy of your voice. WhisprDesk adds no account and no telemetry.

Does WhisprDesk require a subscription?

No. WhisprDesk is a one-time purchase with no subscription, and there is a 7-day free trial with no card required. Only the optional AI cleanup uses your own API key, billed by your provider.

What file sizes and formats can each handle?

Local transcription accepts mp3, wav, m4a, webm, ogg, flac, aac, mp4, wma and opus with no size cap beyond your disk. WhisprDesk's optional cloud mode narrows that to mp3, wav, m4a, webm, ogg, flac and aac, with a 25MB file cap. Long recordings are often easier locally for that reason.

How much disk does local cost me?

Only the models you choose to keep. Whisper ranges from about 75MB for the smallest to about 3GB for the largest, and NVIDIA Parakeet is about 680MB. They download on first use and cache under ~/.cache/whisprdesk/; the installer itself carries no model weights.

Does local dictation need an internet connection?

Dictation and transcription run on-device, so they work without a connection once the app is installed. Only the optional AI cleanup step contacts an external provider.

Local vs cloud dictation: what is the difference?

The difference between local and cloud dictation is where your voice is processed. Cloud dictation sends your audio to a remote server, which usually means an account, an internet connection, and an ongoing subscription tied to usage. Local, on-device dictation does the speech-to-text on your own machine, so your voice never leaves it: no upload, no account, and nothing metered per minute. WhisprDesk is local-first: live dictation and file transcription run on your Mac using Whisper or NVIDIA Parakeet. The trade is that on-device work uses your Mac's own compute rather than a data center, which is why WhisprDesk targets Apple Silicon. If you want optional AI cleanup of the text, you bring your own API key, so that step is your choice, not a built-in cloud dependency.

Where your voice goes

With cloud dictation, your audio is uploaded to a provider's servers, transcribed there, and the text is sent back. Your speech passes through, and is often stored on, infrastructure you do not control.

With local dictation, the model runs on your Mac. The audio stays on the device, so there is no upload and no third party holding your recordings. WhisprDesk works this way for both live dictation and audio-file transcription.

What each costs

Cloud dictation is usually a subscription, with the price tied to how much you use it. Local dictation has no per-minute cloud bill because the work happens on your own hardware.

WhisprDesk is a one-time purchase, not a subscription: pay once and use it. There is a 7-day free trial with no card required. The only usage-based cost is optional: if you turn on AI cleanup, that uses your own API key and is billed by your AI provider, not by WhisprDesk.

When cloud makes sense, when local wins

Cloud can make sense when you need to dictate from a low-powered device that cannot run a model itself. Local wins when privacy matters, when you want predictable one-time cost, or when you need dictation that keeps working without a connection.

The differences that actually bite

Most comparisons stop at privacy. In daily use the practical differences show up elsewhere, and they cut both ways.

File size and format. Local transcription accepts mp3, wav, m4a, webm, ogg, flac, aac, mp4, wma and opus, with no size cap beyond your own disk. Cloud services typically impose both; in WhisprDesk's own optional cloud mode the accepted list narrows to mp3, wav, m4a, webm, ogg, flac and aac, and files are capped at 25MB. A long meeting recording can exceed that while being trivial to process locally.

Setup cost. Local means a one-time download of the model you choose, from about 75MB for the smallest Whisper up to about 3GB for the largest, cached on your Mac. Cloud means no download and no disk use, but a key, an account somewhere, and a per-minute bill.

Failure modes. A cloud service can be slow, rate-limited, or simply down, and you wait. A local model cannot be rate-limited, but it is bounded by your own machine, and a larger model on a busy Mac is slower than a small one.

Neither is automatically more private

Local processing means the audio never leaves your machine, which removes a whole category of questions. But precision matters more than the slogan.

What makes WhisprDesk's default private is not the word local, it is that transcription runs on a model on your disk and nothing is transmitted to do it. The app has no telemetry and no account. If you deliberately switch transcription to a cloud model with your own key, audio does leave, and that is a real change you are choosing, not a hidden one.

Equally, a licensed copy does re-check its licence periodically, so this is not a tool designed for a permanently disconnected machine. Ordinary offline use, a flight or a train, works fine.

How WhisprDesk does it

WhisprDesk runs the speech model on-device using Whisper or NVIDIA Parakeet, your choice. Because it runs locally, it targets Apple Silicon Macs on macOS 12 and later. No account, no telemetry, and audio never leaves your Mac.

Frequently asked questions

Is local dictation more private than cloud?: Yes. With on-device dictation your audio is processed on your own Mac and never uploaded, so there is no server-side copy of your voice. WhisprDesk adds no account and no telemetry.
Does WhisprDesk require a subscription?: No. WhisprDesk is a one-time purchase with no subscription, and there is a 7-day free trial with no card required. Only the optional AI cleanup uses your own API key, billed by your provider.
What file sizes and formats can each handle?: Local transcription accepts mp3, wav, m4a, webm, ogg, flac, aac, mp4, wma and opus with no size cap beyond your disk. WhisprDesk's optional cloud mode narrows that to mp3, wav, m4a, webm, ogg, flac and aac, with a 25MB file cap. Long recordings are often easier locally for that reason.
How much disk does local cost me?: Only the models you choose to keep. Whisper ranges from about 75MB for the smallest to about 3GB for the largest, and NVIDIA Parakeet is about 680MB. They download on first use and cache under ~/.cache/whisprdesk/; the installer itself carries no model weights.
Does local dictation need an internet connection?: Dictation and transcription run on-device, so they work without a connection once the app is installed. Only the optional AI cleanup step contacts an external provider.