Local Voice — Privacy Policy
Last updated: 8 October 2026
This policy describes Local Voice version 1.0.3, currently named “Local Dictation” inside the extension. Local Voice converts speech into text locally in Chrome. It does not use a cloud transcription service or send microphone audio, transcripts or website field contents to the developer.
Microphone audio and transcription
The extension accesses your microphone after you grant permission and start dictation. It also briefly accesses the microphone when you explicitly request microphone permission in its setup or settings pages. Audio is processed on your device using downloaded speech recognition models. Recording stops before transcription begins, and each recording is limited to one minute.
Audio samples are held temporarily in memory for processing; the extension does not save audio recordings to browser storage or upload them. Audio buffers are released after processing. Speech recognition runs locally using cached models, without contacting a transcription service or model server.
Website fields and text insertion
Content scripts run on supported HTTP and HTTPS pages and their frames to track the focused editable field and selection. To insert text safely and detect changes while you dictate, the extension temporarily holds the field’s existing text or editable HTML and selection in page memory. This handling can occur when you focus or interact with an editable field, before you start dictation. Standard password inputs are excluded.
Field contents and selections are not sent to the developer or a remote service by the extension. The extension does not maintain a browsing-history log. Temporary field snapshots are replaced as you interact with fields and are discarded when the page is unloaded.
Your transcript is delivered to the selected website field. That website can then process, save or transmit it under its own privacy policy, just as it can with text you type, including through automatic saving.
What stays in your browser
- Models: downloaded model files and configuration are kept in Chrome Cache Storage so transcription can work offline.
- Preferences: Chrome local storage keeps your selected model, installed-model status and versions, processing mode, onboarding completion and last model used.
- Temporary session data: Chrome session storage holds operating state, tab and frame identifiers used for insertion, errors, and local performance information such as recording duration and transcription time.
- Latest transcript: the most recent transcript is kept in session storage for recovery, including after successful insertion. It is cleared when the next dictation reaches the model-loading stage or when the browser session ends. A failed attempt to start dictation may leave the previous transcript available.
The extension does not use Chrome sync storage or sync this data to a developer account. If insertion fails, you can explicitly copy the recovery transcript to your system clipboard. Your operating system or other clipboard software may retain or sync copied text under its own settings.
Model downloads and network requests
When you choose to download, update or re-download a model, the extension requests model files from Hugging Face and its file-delivery providers. Download requests contain model file addresses, not audio, transcripts or website field contents, and are made without credentials or a referrer.
File providers receive ordinary connection information needed to serve the download, including your IP address and the requested file address, and may handle it under their own policies. See Hugging Face’s privacy policy. The developer does not receive your audio or transcripts through these downloads.
JavaScript and WebAssembly runtime files are packaged with the extension. Internet access is needed for model downloads and Chrome’s extension installation and update mechanisms. Once the required models are cached, transcription itself does not require internet access.
Accounts, analytics and sharing
The current extension has no user accounts, advertising, analytics, tracking, remote crash reporting or developer backend. The developer does not collect or sell your audio, transcripts, field contents or browsing history through the extension. Local processing and temporary storage are used to provide dictation, insertion, recovery and local status information.
Your controls and deletion
You can stop dictation using the extension controls or its keyboard shortcut, and revoke microphone access through Chrome or your operating system’s privacy settings. You can remove downloaded models in the extension’s Settings. Ending the browser session clears temporary session storage; removing the extension removes its extension-managed storage. Text already inserted into websites or copied to the clipboard is controlled separately by those websites or your clipboard software.
Visiting this policy page
This page is hosted through GitHub Pages and uses optional Google Analytics only after you accept analytics in the website cookie banner. Website analytics is separate from the extension and does not receive microphone audio, transcripts or website field contents from it. See the website privacy and cookie policy for hosting, analytics and local storage details. Use “Cookie settings” to change your choice.
Changes and contact
This policy will be updated when the extension’s data handling changes. The latest version is published at this address. For developer information and contact links, visit webh.dev.