Transcribe. Understand. Improve.
Spraakzaak is a modular system for processing, organizing, and evaluating meeting transcripts, designed for public sector organizations.
The goal is to make meetings more accessible through automatic transcription, speaker recognition, and summarization, with focus on privacy, modularity, and scalability.
Spraakzaak has three main components, each maintained in a separate repository:
A web application for uploading or recording meetings, scheduling transcriptions, and viewing or summarizing meetings.
Repository: spraakzaak-frontend
Handles the core processing tasks:
- Speech-to-text transcription
- Diarisation (speaker recognition)
- Meeting summarization
It supports multiple workers for horizontal scalability, models (ASR, diarisation, summarization, validation), and can be deployed on-premise.
Repository: spraakzaak-backend
Used to collect performance metrics such as:
- WER – Word Error Rate
- DER – Diarisation Error Rate
These metrics are based on human-annotated meetings and are used to evaluate and improve the system’s accuracy.
Repository: spraakzaak-evaluator
This is a public project, but not open source. Thus, participation or contributions are subject to approval by the project partners:
- Gemeente Amsterdam (City of Amsterdam)
- Responsible IT Lab, Amsterdam University of Applied Sciences (HvA)
For questions or participation inquiries:
-
Marcio Fückner – Researcher and Lecturer
✉️ m.fuckner@hva.nl -
Virgill van der Meer – Gemeente Amsterdam
✉️ v.van.der.meer@amsterdam.nl