On-Device Transcription on Mac: How Local Models Work

On-device transcription means a speech-recognition model runs on your Mac and turns audio into text there. In Avaan this is the free local transcription on Mac: no account, no minute allowance, and the audio stays on the Mac. The tradeoff is that you do the computing, you download a model, and you get fewer features than Pro cloud transcription, which is the main experience in Avaan.

This guide covers how local models work and when each approach fits. If your question is specifically what works with no connection, read offline transcription on Mac.

How a local speech model runs

A speech model is a neural network trained to map audio to text. Running it locally has four steps:

1. The app captures audio from the microphone or opens an audio file.
2. It splits the audio into short windows and feeds them to the model.
3. The model predicts words, and the app assembles them into a transcript.
4. The text lands where you were typing, or in a file transcription result.

Avaan's local transcription on Mac uses a Parakeet model by default. The model is a file on your disk, which is why the first use involves a download, and why later use does not need a network.

The model's work happens on your Mac's processor, so speed depends on your hardware, the length of the recording, and what else the Mac is doing. Short dictation feels immediate on most recent Macs. A long file takes longer, and you can keep working while it runs.

What local transcription gives you

• No account. You can use it right after installing.
• No allowance to watch. There is no monthly minute budget on local transcription.
• Offline use. After the one-time model download, it keeps working without a connection.
• Audio stays on the Mac. Local transcription does not upload that audio.

That last point is narrower than it sounds. It describes where the audio goes during transcription. It says nothing about what you do next: if you paste the text into a cloud document or email it, that text goes where you send it.

What it costs you

• A model download and some disk space. Check the model's size in the app before you download on a metered connection.
• Your Mac's resources. Long files compete with your other work for processor time.
• Fewer features. Modes, speaker labels, custom vocabulary sync, and Ask are Pro features that run in the cloud.
• Accuracy that depends on the audio. Local models handle clear speech well. Noise, crosstalk, accents, and rare terms are where cloud models are usually the better bet. Results vary by recording, so compare on your own audio.

How cloud transcription differs

Pro cloud transcription sends audio to speech providers, which return text in real time on iPhone and Mac. The subprocessors page lists the providers and the privacy policy describes what Avaan stores. Dictation history syncs metadata only, never transcript text.

Cloud gives you the features that depend on more than a speech model. Modes format the text for the app you write in, speaker labels mark who spoke, and on iPhone, Notes sync to your account with memories you can turn on to Ask questions across them.

Pro is $7.99 a month or $79.99 a year, with 10 hours of cloud transcription each month. Every account gets a 7-day trial with no card, up to 140 minutes of cloud transcription. After a trial or a cancelled plan, cloud features pause, Notes stay readable, and local transcription on Mac stays free. See pricing.

Choosing per job

Apple's built-in Dictation belongs in this comparison too. Apple's support page says you can check Keyboard settings to see whether general text Dictation is processed on your device. If it covers your language and your use, it costs nothing and needs nothing installed, and it is a perfectly good choice for short text.

A simple way to decide

Start local. Download Avaan, dictate and transcribe a few of your own recordings, and note where the text needs more correction than you want to do by hand. Run those same recordings through the free Pro trial. If the difference matters for your work, Pro is worth it; if not, the free local tier does the job.

For how dictation and file transcription differ as workflows, see real-time dictation vs file transcription.