What the Mac handles locally
The local recognition model turns available audio into words on your Mac. Once the models are prepared, this step does not require sending the recording to a cloud recognizer. Local speaker identification also runs on the device and needs its additional model downloaded first.
Whynote has separate provider settings for note transcription and voice dictation. Check both if you want local recognition in both workflows. The offline setup guide covers model downloads and the controls.
What depends on other features
| Operation | Where data is processed |
|---|---|
| Local speech recognition | A model on the Mac converts audio into text. |
| Local speaker identification | An additional model on the Mac separates speech into labels. |
| Cloud transcription | Audio is sent to the service for recognition. |
| LLM text enhancement | Recognized text is sent for online processing when the feature is enabled. |
| Note summaries and chat | The relevant content is used by server-based AI features. |
| Account synchronization | Notes, transcripts and associated information can be transferred to your account. |
Choosing a local model describes the speech-recognition step. When you use an account, text can still sync even after local recognition with enhancement turned off. Consider the recognition setting together with the other features you use.
Choose the processing steps you need
- Download the local model in Settings → Transcription.
- Choose Local (On-Device) for Note Transcription Provider or Dictation Provider.
- Check LLM Text Enhancement in the relevant section and turn it off if you do not want the additional online rewrite.
- Prepare the local diarization model if you need speaker labels.
- Review account use and synchronization separately from those controls.
You can use local speech recognition for a transcript and edit the note yourself afterward. If you later request an AI summary or chat response, the relevant text is used for that online feature. Choose those steps deliberately for the material you are working with.
Consider storage and the finished text
Check where you keep the source recording and who has access to that folder. Copied text may end up in email, a document or a work chat with its own synchronization. On-device recognition does not determine where you send the result afterward.
Remove details the recipient does not need and check the destination before sharing. If a project requires no data transfer at all, the local-recognition setting alone does not establish that condition. Account behavior and every feature involved need their own verification.
When to choose local or cloud recognition
Local recognition is useful when you want the recognition step on your Mac or need a prepared model without internet. Cloud recognition needs a connection and sends the audio to the service. Check language support and a representative recording with the provider you choose.
Your available transcription allowance still applies to local processing, as shown in the app. Connected AI features have their own Pro requirements. See the privacy policy for data handling and the audio-file guide for import steps.
Try your chosen workflow in Whynote
Get Whynote for Mac to prepare a model and choose how to process your recording. Use the iPhone app to keep a voice note or import audio saved on your phone.