- Does TALDEN really run fully offline?
- Yes. After installation — which downloads AI models once — no network connections are required or initiated. Verifiable via firewall log or netstat during an active session.
- How good is the recognition quality for German?
- TALDEN uses current Whisper models at the state of the art for German. Quality varies with audio quality — microphone, ambient noise, speaking speed. For critical applications, we recommend a pilot to validate in your environment.
- How is TALDEN deployed?
- TALDEN is provided as a Windows installer (MSI package). MDM-compatible, silent install via msiexec, administrator documentation on request.
- What happens when AI model updates are released?
- Models are delivered via regular application updates. Each update is tested before release. Organisations with approval processes can manage their own update paths.
- How are speakers distinguished?
- pyannote separates speakers locally and assigns labels (Speaker A, B, C). Names are not identified — mapping to individuals is the responsibility of the user.
- Can we inspect the source code?
- Architecture documentation including a dependency overview is available within pilot projects. Full source code access is subject to individual agreements.
- Does the VS-NfD certification also apply to TALDEN itself?
- No. TALDEN is VS-NfD-ready: the architecture meets the technical prerequisites for deployment in VS-NfD-accredited environments. A product-specific BSI Zulassung is in preparation — TALDEN is not separately certified.