Murmel

Details

Name of the digital solution

Murmel

Description

Murmel is a Dutch-optimized speech-to-text service developed by The AI Factory in Amsterdam. Most commercial speech recognition models are trained primarily on English and process audio on servers outside the EU; Murmel is trained specifically on thousands of hours of Dutch audio and runs entirely on infrastructure in the Netherlands. The model achieves a 1.6% character error rate on Dutch audio, outperforming Azure, Mistral, and leading open-source alternatives — including on accented speech and domain vocabulary used in government, healthcare, and media. Processing runs at roughly 35× real-time: one hour of audio transcribes in about two minutes. Murmel is built as a sovereign-by-design alternative for organizations that cannot send sensitive audio outside the EU. All processing takes place on Dutch servers hosted with LeafCloud, a sustainable Amsterdam-based cloud provider. Customer audio is never used to train models. Organizations can choose in-transit-only or do-not-save modes depending on their retention requirements, and a standard DPA is included to simplify AVG/GDPR compliance. The service is offered through a web application (Murmel Studio) and a REST API for integration into existing workflows. Use cases include transcription of council meetings, parliamentary and court audio, medical consultations, journalistic interviews, and media captioning — settings where Dutch accuracy and EU data residency are mandatory rather than optional. Pricing is transparent and usage-based: a free tier (60 minutes/month), subscription tiers up to 100 hours/month, pay-as-you-go credits, and enterprise SLAs. Support is delivered from Amsterdam in Dutch, in the same time zone as customers. Murmel gives European organizations a trusted, locally-hosted alternative to US and Chinese speech recognition services without compromising on accuracy.