Dutch speech recognition that actually understands Dutch.
Generic speech models make errors on Dutch audio and process your data on servers outside the EU. Murmel is trained specifically on thousands of hours of Dutch, fully hosted in the Netherlands, and GDPR-compliant by design.
Built in the Netherlands · Hosted in the Netherlands · Best for Dutch
Sign up now →First users are testing! Sign up now or contact us to help test.

Suitable for government, healthcare and media
Most speech models are trained primarily on English. On Dutch audio they fall short — and they process your data on servers outside the EU.
All processing on Dutch servers — GDPR-compliant by design, with no dependence on Big Tech providers.
Developed by The AI Factory in Amsterdam. Direct support in Dutch and English, in the same time zone.
Built for sensitive data: data stays in the Netherlands, in-transit or do-not-save options, and no AI training on your audio.
1.6% CER on Dutch audio — trained on thousands of hours of Dutch, better than Azure, Mistral and open-source models.
Whether you transcribe an occasional interview or process hours of audio daily-there is a suitable package for everyone.
60 free credits included — no credit card required. Better accuracy on Dutch than Azure, Mistral or open-source — tested on thousands of hours of Dutch.
Read the benchmark results