Public organisations join forces for Dutch speech recognition

Speech is increasingly becoming the natural interface between people and technology. Think of customer service, government services, healthcare institutions and digital assistants. Automatic transcription, subtitling and voice-controlled applications are also becoming more important in education and research.

Speech models are often developed outside the Netherlands, which means a great deal of post-processing is required. With HOSAN (High-Quality and Sovereign Speech Recognition for All Dutch), participating organisations are taking their first joint step towards a future in which speech is an essential basic service. High-quality speech recognition for all varieties of Dutch will not only become available, but can also be developed and managed in-house.

The launch of HOSAN

On Thursday 25 June, the HOSAN project was launched during the conference ‘Samen versnellen met AI: de overheid van morgen begint nu’. Around 700 administrators, policymakers, researchers, implementing organisations and technology partners took part. By signing a letter of intent, the participating organisations expressed their support during the conference. The shared ambition: to achieve high-quality, reliable and digitally sovereign speech recognition for all forms of Dutch, including accents, dialects and different speaking styles.

From exploration to development

In recent years, public organisations, research institutions and civil society partners have jointly investigated how high-quality and sovereign speech recognition for the Dutch language can be achieved. They have examined technological possibilities, available data, infrastructure, governance, legal frameworks and societal applications. SURF’s High Performance Machine Learning (HPML) team has so far played a role in assessing the technical feasibility of the speech model, for which Snellius has been utilised. This has laid the foundations for actually realising this ambition.

Stakeholders

In addition to SURF, the stakeholders include: Beeld & Geluid, the NPO, the National Police, the Ministry of Justice and Security, the Ministry of the Interior and Kingdom Relations, the Ministry of Economic Affairs and Climate Policy, the University of Twente, Radboud University, DUO, Amsterdam University of Applied Sciences, the Royal Netherlands Marechaussee, Kennisnet, the General Secretariat of the Taalunie, ICTU, the Alliance for Digital Coexistence and the AI & Government Platform.

Eppo van Nispen tot Sevenaer, director of Beeld & Geluid and one of the initiators of the partnership: “In recent years, we have been investigating whether high-quality Dutch speech recognition is possible. The answer is clear: yes. We have the expertise, the data and the partners to make this a reality. With HOSAN, we are investing not only in better technology, but also in our digital autonomy and the preservation of our language and culture in the AI era.”

Would you like more information about HOSAN?

Visit www.hosan.nl. There you’ll also find a list of frequently asked questions and answers.