How local recognition works in Whynote
After you download the model, Whynote can recognize speech on your Mac without sending the audio to a cloud recognizer. This is useful when internet is unavailable or you prefer to process the recording on your device. Prepare it while you still have a connection to download the model.
Summaries, chat and text enhancement use online processing. Before working offline, check the recognition provider and turn off enhancement for the workflow you intend to use. The steps below cover notes and dictation separately.
You can also prepare speaker diarization locally with an additional model. Account transfer of notes and text remains a separate part of the workflow. The local processing and privacy guide explains why turning off text enhancement does not turn off account synchronization.
Download the model and choose the correct provider
The current Mac download requires Apple silicon and macOS 14 or later. Install the official app in Applications and open Settings → Transcription. In Local Model (FluidAudio), use Download and wait until the model is ready. If a download was interrupted and the app offers Resume, use that control while a connection is available.
- For meeting notes: choose Local (On-Device) under Note Transcription Provider.
- For voice dictation: choose Local (On-Device) under Dictation Provider.
- For offline text processing: switch off LLM Text Enhancement in the relevant note or dictation section. These are separate controls.
- For local speaker identification: download the separate model in Speaker Diarization, then enable Enable Speaker Identification when it is ready.
- Before relying on the setup: process a short sample and read the result. Make sure the sample includes the language and audio source you expect to use.
Note enhancement can send locally recognized text to the server for cleanup when enabled. Dictation enhancement also adds an online step. Turning off one does not turn off the other. Check both if you use both kinds of capture.
If you use both notes and dictation, check both provider settings before disconnecting. Prepare the recording permissions and enough disk space for the audio alongside the model.
Which parts need an internet connection?
| Feature or step | Offline after preparation? | What to check |
|---|---|---|
| Record microphone audio | Local recording | Microphone permission and selected input |
| Recognize recorded speech locally | Yes, with ready models | Local provider, supported language, model readiness |
| Local speaker identification | Yes, with its additional ready model | Diarization enabled; inspect labels afterward |
| Cloud transcription | No | Connection and account access |
| LLM text enhancement | No | Disable it for offline text processing |
| AI summaries, chat and workflows | No | Connection and Pro access |
| Account setup, model downloads and sync | Connected operations | Prepare before leaving a connection |
A local transcript and an AI summary are different outputs. You may produce the transcript offline, edit a useful note yourself and decide later whether to use a connected summary. Likewise, choosing local dictation does not alter the note provider, and choosing the note provider does not alter dictation.
Local processing still uses your Mac's memory, processor and storage. Leave room for the models and recordings, and avoid starting several large processing jobs when you need the computer for other work. Processing time depends on your Mac, the recording length and the current workload.
Check language support before the real recording
Local and cloud providers have different language coverage. The app's local language settings include English, Russian and other supported European languages, but a cloud language option should not be assumed to work with the local model. Choose from the languages available for the selected provider and test a representative sample.
A sample should include the kind of speech that matters to you: an interview, a technical discussion or a lecture with proper names. Check those names manually. Background noise, overlapping voices and switching languages can make recognition harder. Clear audio remains useful regardless of where recognition runs.
Speaker diarization separates portions of speech into speaker labels. It does not know who a person is just because the model assigns a number. Review changes in speaker and correct labels before using them to attribute a decision. If you know the number of speakers, the app offers an expected-speaker setting as a hint; it does not replace that review.
Check the audio before recording
A ready speech model cannot repair a missing audio source. For voice capture, check Settings → Permissions → Microphone. For a call playing on the Mac, also check System Audio, or Screen & System Audio on an older supported macOS version. Check the input device in Settings → Audio after changing headphones or connecting a microphone.
Run a short recording test and listen to it before an important meeting. The meeting recording guide covers microphone and system audio together. For text insertion into another app, follow the dictation setup.
Keep your original recording until you have reviewed the transcript. If a word is uncertain, replay that passage and write the correction. For a useful offline note, you can summarize decisions and questions yourself. A connected AI feature is optional; checking the original remains necessary either way.
Local unavailable, failed or waiting?
| Symptom | Check | Action |
|---|---|---|
| Local is unavailable | Recognition model state | Reconnect to finish or resume the download, then wait until the model is ready. |
| Language is unsupported | Language choices for the Local provider | Choose a supported language and test a representative sample. |
| Local text appears, then an online step fails | Note or dictation LLM enhancement | Disable the relevant enhancement for offline text processing. |
| Speaker labels are unavailable | Separate diarization model and enabled setting | Download that model while connected, then enable speaker identification. |
| Recording waits or captures silence | Microphone permission, input and readiness | Check the audio device; use the offered retry or cancel control and test the recording. |
A partially downloaded model and one still loading into memory are different from a ready model. If you delete the model to recover disk space, prepare it again before your next offline session.
An online error after a local transcript appears can belong to enhancement rather than recognition. Check note and dictation enhancement separately: switching off one does not switch off the other.
If a recording is waiting for the microphone, give initialization time and follow the app's retry or cancel controls. If there is no useful audio, fix microphone access and input selection. Preserve recoverable files until you have a checked result, and use the support contact if a repeatable issue remains.
Offline mode and subscription limits
Local is a processing choice, not a separate unlimited plan. Free has a monthly transcription allowance and limited access to older notes. Open Settings → Subscription to see the current allowance and plan on your account. Sign in and check your available allowance while you still have a connection.
Pro provides the plan's expanded transcription access, older note access and connected AI tools. It does not make AI summaries or text enhancement work offline. Review the current price, billing period and renewal terms in the purchase screen, and choose a plan based on the workflow you intend to use.
Common questions
Does Local mean the whole app is offline?
Local describes speech recognition rather than the whole app. Model downloads, cloud transcription, AI enhancement, summaries and synchronization are connected operations.
Do I need two models to transcribe?
The recognition model is needed for local speech to text. Speaker identification uses an additional diarization model; download it only when you need that feature.
Why does a locally recognized note contact the server?
LLM Text Enhancement is a separate online step. Switch off the note enhancement control for offline text processing, and check the dictation control separately.
Settings labels can vary between app versions. Before an important recording, check your installed version, selected provider and the result of a short test.
Continue this topic
Bring your next recording into Whynote
Get Whynote for Mac to record a conversation or open a saved audio file. Choose the iPhone app for voice notes and recordings you keep on your phone.