AI4Interviews – Innovation in the Police

At the core of the project lies speech-to-text technology, converting interview recordings and observations into written text within minutes rather than hours.

The system supports multiple languages, including Norwegian dialects, and is used for interview transcripts, summary reports, and court transcriptions.

The solution benevis.ai, originally developed for NRK, has been further adapted to meet the needs of the police.

Speech-to-text is also being tested for forensic dictation and in fieldwork (PPS) conducted by patrol officers.

Making Police Interviews More Efficient

The police conduct more than 150,000 interviews annually, of which around 45,000 take place in the field. Manual transcription and report writing have traditionally been highly time-consuming. With AI4Interviews, documentation time can be reduced by up to 80 percent.

This frees investigators to spend more time in the field and focus on high-quality police work.

The project also supports the KREATIV interview method, emphasizing due process, empathy, and scientific integrity.

New Solutions for Crime Scene Investigation

Forensic technicians can use speech-to-text and voice-controlled head-mounted cameras (such as JodaPro) to document findings directly at the crime scene. These cameras can live-stream to experts who are not physically present, and recordings can be automatically transcribed.

Documentation is then structured into standardized reports within police systems (BL, Mediabanken).

This creates a seamless chain from recording → transcription → final report.

Robotic Process Automation (RPA)

Digital automation can reduce repetitive work such as transcription, report structuring, and text analysis. For example, multiple audio files can be consolidated into a single report automatically.

Artificial Intelligence

AI is applied in speech recognition, summarization, and text analysis.

• Named Entity Recognition (NER) extracts names, places, times, and objects from large volumes of text.
• Large Language Models (LLMs) are tested for summarization, comparison notes, and pattern recognition in interviews.
• Custom Norwegian datasets are being built, improving performance on dialects and complex legal language.
• Models such as Whisper and Wav2Vec2 have shown strong results for Norwegian speech-to-text.

Equipment

• Voice-controlled head-mounted cameras (e.g., JodaPro, adapted from healthcare)
• Mobile app for patrol officers (Capture, PPS fieldwork)
• PC interface for interviews and courtrooms
• Integration with secure police systems (BL, PPS, Mediabanken)

AI4Interviews follows a user-centered and iterative innovation approach, developing solutions step by step through pilots and proof-of-concepts (PoCs). Investigators, forensic specialists, lawyers, and technologists are closely involved throughout.

Work is organized into four work packages:

1. User Interfaces – developed with the police, focused on simplicity and efficiency.
2. Speech-to-Text & Text Analysis – research and development of algorithms, language models, and NER.
3. Technical Infrastructure – secure operation within police networks and systems.
4. Innovation & Project Management – coordination, ethics, legal frameworks, and benefits realization.

The project applies a triple-helix model, systematically combining expertise from police, academia, and industry. Continuous testing in real-world settings (interviews, courtrooms, crime scenes) provides feedback for rapid improvement.

Risk and ethics management are integral to the process, with strict adherence to GDPR and the EU AI Act. Solutions are tested first with synthetic or limited data before operational use.

The main goal is to ensure that technology not only works technically but also gains user acceptance and strengthens the police mission—without compromising rule of law.

The project collaborates with:

• NTNU CCIS (language technology, security)
• University of Agder
• Netherlands Forensic Institute
• London Metropolitan Police

This interdisciplinary approach ensures international pace and quality in development.

AI4Interviews demonstrates how new technology can:

• Reduce manual workload
• Increase accuracy and legal certainty in police work
• Strengthen forensic documentation
• Prepare police forces for future challenges

The project has already gained international visibility and is one of the most high-profile innovation initiatives within the Norwegian Police.

Want to know more about the project? Please contact: AI4Interviews@politiet.no

The project is led by Oslo Police District, with broad participation from police units and partners.

• Bente Skattør: Project Manager AI4Interviews, Oslo Police District / Assoc. Professor II, NTNU CCIS
• Grete Lien Metlid – Head of Investigation & Intelligence, Oslo Police District, Chair of the Steering Committee
• Thomas Ibsa Beka – Police IT Unit, responsible for machine learning and technical development