With a downloaded local model, Voicetypr transcribes recorded audio on your device rather than sending it to a hosted speech provider. License checks, updates, telemetry, optional cloud speech, text formatting, LAN sharing, local history, and the destination app remain separate data paths to review.
Last verified: July 27, 2026
Publisher and method: Voicetypr publishes this architecture explainer. Claims were checked against the public product, remote, cloud-provider, licensing, and telemetry source. This is not an independent privacy or security audit.
Quick verdict
Choose a downloaded local model when keeping the speech-recognition step on the device is the main requirement. Do not treat that choice as proof that the whole app is zero-collection, zero-network, compliant, or air-gapped; review licensing, downloads, telemetry, optional providers, LAN sharing, local history, and the destination app separately.
Retention vs. collection
Processing location, retention, telemetry, and the destination app are separate questions. Here is the decision boundary this page can support from public source evidence.
Audio leaves the device for speech recognition. Retention, training, access, and deletion depend on the selected provider and plan; verify its current terms.
A downloaded model handles the recorded audio on the same computer. That narrows the speech data flow, but it does not erase licensing, updates, telemetry, local history, or the destination app.
Cloud speech sends audio, AI formatting sends text, and LAN sharing sends audio to another trusted device. Choose deliberately and assess the receiving system.
A service can process speech remotely without retaining it indefinitely, or store data under a stated policy. Check each provider's current documentation instead of assuming every cloud service has the same retention behavior.
Local speech recognition keeps that recognition step on-device, but the resulting text is still inserted into another app. That app, its sync settings, backups, and collaborators can govern what happens next.
Cloud speech sends audio to the chosen provider, AI formatting sends text, LAN transcription sends audio to another machine, and telemetry can send scrubbed operational events. Review the active configuration, not only the default model label.
In local mode, your microphone captures audio and local AI models (Whisper and Parakeet) process it on your device. The audio file is not transmitted to Voicetypr servers or any cloud transcription service.
The transcript is generated locally and inserted into your destination app. Voicetypr's public telemetry code is designed to reject arbitrary transcript content, but local history, the destination app, sync, and backups remain part of your environment.
License or trial checks, updates, telemetry, support actions, optional cloud speech, AI formatting, and LAN sharing are separate from local model inference. Each should be reviewed on its own terms.
Local speech recognition can keep recorded audio off a hosted speech provider. The destination editor, local history, sync, backups, and organizational policies still need review.
Local transcription can reduce one third-party audio transfer. It does not by itself establish privilege, confidentiality, compliance, or suitability for a firm's approved systems.
Use local models when hosted speech processing is outside the approved workflow, then evaluate the destination app, storage, collaboration, and backup path separately.
Feature-by-feature boundary
The correct verdict depends on the selected model, settings, and destination. A local model narrows the speech-recognition path; it does not certify a workflow or remove every other network and storage path.
Use a downloaded Whisper or Parakeet model when your requirement is to avoid sending recorded audio to a hosted speech provider.
Review that provider's privacy, retention, security, and contractual terms before enabling it for sensitive material.
The audio can remain local while the generated text leaves the device. Treat formatting as a separate disclosure and risk decision.
The request stays between your machines rather than a hosted speech provider, but the server and local network become part of the trusted boundary.
Local speech recognition is not a promise that the entire app makes no network requests or collects no operational data.
This page is a product data-flow explanation, not legal advice, a certification, or a guarantee of compliance. Organizations remain responsible for device security, approved vendors, contracts, retention, access, backups, and the destination application.
Direct answers based on the currently documented product and public source boundaries.
related pages
Compare local, LAN, disconnected, and policy-level boundaries before choosing a workflow.
Healthcare
Review the product limits and your own healthcare-compliance responsibilities before using dictation with sensitive data.
Offline
Works offline by default. Audio stays on your machine during transcription.
Compliance
Review the local-processing boundary alongside your own GDPR roles and obligations.
Product
Private offline AI voice dictation for Mac and Windows. No subscription.
3-day free trial. No credit card. Use a downloaded local model when on-device speech recognition is your priority. Pay once, use forever.