Transcription for government

Dutch speech recognition on Dutch servers: digitally sovereign, GDPR-compliant and publicly benchmarked. For council meetings, hearings and interviews.

100% hosted in the Netherlands · no US cloud

By Maarten Sukel, founder of Murmel. Updated on 28 October 2026.

Digital sovereignty beyond the policy paper

Governments want to depend less on American cloud services, yet recordings of meetings and conversations often still pass through tools that process audio outside the EU. Murmel is the opposite route: a Dutch speech recognition model, trained by The AI Factory, running on servers in the Netherlands. No US cloud is involved, from upload to deletion.

  • Processing entirely on servers in the Netherlands; data never leaves the EU.
  • You decide how long recordings and transcripts are kept, with automatic deletion from 1 day.
  • Data is not used for model training unless explicitly chosen.

Tested on exactly this kind of speech

The quality is not just a claim: the model is publicly benchmarked on 8.9 hours of Dutch parliamentary debate, with speakers from every province, and the current version makes the fewest errors of seven tested models on four open Dutch test sets (Murmel v5). Meeting language, interruptions and policy jargon are precisely the domain it is trained and tested on; add your own terminology via a custom vocabulary.

What governments use Murmel for

  • Council and committee meetings: from recording to searchable report with speaker labels, as the basis for minutes or decision lists.
  • Hearings and conversations: word-for-word reporting where required, verifiable with the audio next to the text.
  • Interviews for policy research: transcripts that fit a data management plan, see also transcription for research.

Curious about the broader movement towards European digital infrastructure? Read the piece on digital sovereignty.

Getting started

Try it right away: start with a 14-day trial of Basic, then €10.89 per month (incl. VAT), cancel monthly. For council meetings the workflow is described in transcribing council meetings. For teams, higher volumes, a data processing agreement or API integration: get in touch.

Frequently asked questions

Where is the audio processed?

Entirely on servers in the Netherlands. No American or non-European cloud is used, and data never leaves the EU at any point.

Is a data processing agreement possible?

Yes, a data processing agreement can be arranged for organisations. Get in touch via the contact page.

How well does it handle meeting language and accents?

The model is publicly benchmarked on Dutch parliamentary debates with speakers from every province and the current version makes the fewest errors of seven tested models on open Dutch test sets. The methodology is published and verifiable.

Can Murmel connect to existing systems?

There is an API for submitting recordings and retrieving transcripts programmatically. Get in touch to discuss the integration.

Dutch speech, Dutch servers

Try it yourself or discuss the options for your organisation.

Get in touch