Audio Transcription
Configures automatic speech-to-text transcription for audio notes recorded in the Notepad Quick Entry. When enabled, audio notes are transcribed in the background on the server immediately after they are saved, and the note text is filled in with the result.
Everything runs locally — no internet connection or cloud service is involved.
Status
Section titled “Status”The status card at the top of the page shows:
- Notes waiting — audio notes that have not yet been transcribed and are waiting for a model to become available.
- Notes queued — notes that were queued while transcription was unavailable. Use the Transcribe queued notes button to process them once a model is ready.
- Active model — the model currently loaded by the transcription engine.
- Engine status chip — whether the engine is running, idle, or in an error state.
If the engine encountered an error, the details are shown here. Errors on individual notes are shown as warnings and do not stop the engine.
Speech-to-Text Transcription
Section titled “Speech-to-Text Transcription”Enable Transcription
Section titled “Enable Transcription”Turns automatic transcription on or off. When enabled, every new audio note is transcribed as soon as it is saved. Notes recorded while transcription is disabled are not retroactively transcribed.
Select the speech recognition model from the dropdown. Each model shows its download status inline. Two models are available:
| Model | Languages | Size |
|---|---|---|
| NVIDIA Parakeet TDT 0.6B v3 | 25 European languages with automatic language detection | ~490 MB |
| NVIDIA Parakeet TDT 0.6B v2 | English only — slightly more accurate for English notes than v3 | ~485 MB |
If no model is selected, audio notes are queued and transcribed once a model is selected and downloaded.
Model Files
Section titled “Model Files”When a model is selected, its download status and size appear below the dropdown:
- If the model is not downloaded, a Download button initiates the download directly from the internet to the server. A progress bar shows download and extraction progress. If a model is not yet downloaded, audio notes are queued until the download completes.
- If the model is already downloaded, a Delete button removes it from disk to free space. Deleting the active model disables transcription until a new model is selected and downloaded.
Click Save & Restart Transcription to apply changes. This restarts the transcription engine with the new settings.