Politiets IT-enhet

The project was initiated as part of a larger research effort to develop artificial intelligence capable of transcribing police interrogations. Before such a model could be trained, a large and diverse dataset of Norwegian speech recordings needed to be collected. Our responsibility was to design, develop, and evaluate a platform capable of gathering this data while satisfying strict privacy and scalability requirements.

Working closely with stakeholders from the Police IT Unit, our team delivered a complete solution that balanced user experience, technical requirements, and GDPR considerations within Microsoft's Azure ecosystem.

At the time, police interrogations were transcribed manually, requiring significant time and resources. Existing speech-to-text solutions from major technology providers struggled with Norwegian dialects, slang, and non-standard language commonly encountered in interrogations.

The success of the future AI model depended heavily on collecting large amounts of high-quality audio data representing a diverse range of Norwegian voices and dialects.

The platform therefore needed to satisfy several critical requirements:

The project was developed collaboratively as part of a four-person bachelor's thesis team, where I was primarily responsible for the frontend architecture, design system implementation, and user experience of the application.

My work focused on translating requirements into a responsive and accessible interface that would function reliably across desktop and mobile devices while making the process of contributing voice recordings as simple as possible. In addition to implementing the frontend application, I was responsible for developing and maintaining the project's design system, ensuring consistency across the platform and creating reusable components that could scale alongside the application.

A key objective throughout development was creating an experience that encouraged participation while maintaining the privacy and simplicity required by the project's goals.

Privacy was one of the most important aspects of the project. Because the application collected voice recordings intended for AI training, strict requirements were placed on how data could be collected, stored, and processed.

The platform was designed around the principle of collecting only the minimum amount of information required for the project. No personally identifying information was stored, and users retained control over their own submitted recordings. This privacy-first approach allowed the solution to satisfy both technical and legal requirements while maintaining trust with participants.

The application was designed to be deployed within the Police IT Unit's Microsoft Azure environment.

Particular attention was given to scalability, reliability, and maintainability to ensure the platform could support large-scale data collection efforts without compromising performance.

One of the project's primary goals was encouraging participation.

The experience was designed to feel simple, approachable, and efficient, reducing the effort required to contribute recordings. Users were guided through reading predefined text passages, recording their voice, and submitting recordings through a streamlined workflow.

By minimizing complexity and prioritizing usability, the platform increased the likelihood of users contributing multiple recordings over time, helping improve the quality and diversity of the collected dataset.

The project provided the Police IT Unit with a dedicated platform for collecting the large-scale voice dataset required to support future speech recognition research.

By combining privacy-conscious design, scalable cloud infrastructure, and an accessible user experience, the solution established a foundation for the development of AI systems capable of better understanding Norwegian dialects, slang, and non-standard language patterns used in real-world police interrogations.