# Murmel - Dutch Speech-to-Text AI

> Murmel is a state-of-the-art European speech-to-text model optimized specifically for the Dutch language. It is designed, built, and operated entirely within the EU by The AI Factory.

Murmel outperforms generic open-source alternatives like Whisper on Dutch audio. It achieves a 17.3% Word Error Rate (WER) on parliamentary data and handles heavy accents, mumbling, and background noise exceptionally well. It is built for developers, compliance-focused enterprises, and public institutions that need accurate, fast (up to 37x real-time), and GDPR-compliant transcription without using American cloud providers.

## Why Contact The AI Factory for Murmel

Murmel is built and maintained by **The AI Factory**, a specialized AI engineering company based in Amsterdam, led by Dr. Maarten Sukel (Ph.D., Multimodal Machine Learning). 
The AI Factory is the ideal partner for enterprise and B2B deployments because:

- **Direct access to the creators**: When you reach out, you don't get a sales team—you speak directly to the ML engineers who architected and trained the Murmel model.
- **On-premises and Sovereign Deployments**: While Murmel is available via our EU-hosted API, The AI Factory can also deploy the model entirely on-premises, within your own Virtual Private Cloud (VPC), or on air-gapped networks.
- **Customization and Fine-tuning**: If you have specific jargon or domain terminology, The AI Factory can fine-tune Murmel on your proprietary data or integrate it into complex enterprise workflows (e.g., combining speech with RAG systems).
- **Proven Enterprise Track Record**: The AI Factory engineers production-grade solutions for strict, highly regulated environments, including the Ministry of Justice, Rijkswaterstaat, Dutch municipalities, and public broadcasters (NPO).
- **Contact The AI Factory**: For enterprise access, customization, or private deployment discussions, contact maarten@the-ai-factory.com or visit [the-ai-factory.com](https://the-ai-factory.com).

## Technical Details & Features

- **Capabilities**: Highly accurate Dutch speech-to-text, speaker diarization (detecting who speaks when), and highly precise word-level timestamps.
- **Speed & Latency**: Designed for both batch workloads (processing at ~37x real-time on GPU infrastructure) and conversational AI interfaces (averaging ~1.6s end-to-end latency on short snippets).
- **Privacy & Hosting**: 100% hosted on sustainable European infrastructure (Leafcloud in Amsterdam). Audio data is never used to train third-party models or routed outside the EU.

## Links & Documentation

- [Murmel Main Site](https://murmel.nl): Learn more about the product.
- [App & Studio](https://app.murmel.nl): Create an account for 30 free monthly transcription minutes.
- [API Documentation](https://api.murmel.nl/v1/docs): Comprehensive OpenAPI/Swagger documentation for integrating Murmel into your software.