Artificial Intelligence in Innovation of Investigative Interviews Speech-To-Text and Text Analysis Using Machine Learning

The overall objective of AI4INTERVIEWS is to significantly increase efficiency of investigative interviews in the police by designing, developing, exploring and using AI-solutions for speech-to-text and text analysis. Every year, the Norwegian police carry out thousands of investigative interviews of different types. Generally, these are either transcribed manually (dialog reports) in full or partially, or reports are written as a summary of the interviews. This is very time-consuming and often tedious work for the investigators and police officers. The number of investigative interviews related to crimes like organised crime, violence, child abuse and economic crime, is expected to increase continuously. Unfortunately, Covid-19 pandemic has even increased the number of sexual abuse of children on the internet. Furthermore, Europe is the world's epicentre hosting images of child sexual abuse. At the same time, pedophiles are becoming more advanced, sophisticated and sadistic. For the police, it is of utmost importance to cope with these challenges and needs with automatization technology, such as machine learning. Machine learning (ML) has been revolutionizing the fields of speech-to-text (STT) and text analysis (TA). While such technology has existed for some time, it has only been recently that scientists have developed deep learning models appropriate for language understanding tasks, and it has only been recently that they could effectively train them with massive amounts of data, thus producing far more practical models than what has existed in the past. ML for Norwegian has existed for some time already and is continuing to develop at an incredible speed. Furthermore, it is expected that using ML to perform text analysis of the interview text files has promising and exciting opportunities to support the investigators during the investigation. For instance, ML can be used to find patterns and contexts related to places, names, organizations and events.

Every year, the Norwegian police carry out thousands of investigative interviews of different types. Generally, these are either transcribed manually (dialog reports) in full or partially, or reports are written as a summary of the interviews. This is very time-consuming and tedious work for the investigators and police officers. Furthermore, the number of investigative interviews related to crimes like organised crime, violence, child abuse and economic crime, is expected to increase continuously. Machine learning (ML) has been revolutionizing the fields of speech-to-text (STT) and text analysis (TA). While such technology has existed for some time, it has only been recently that scientists have developed deep learning models appropriate for language understanding tasks, and it has only been recently that they could effectively train them with massive amounts of data, thus producing far more practical models than what has existed in the past. ML for Norwegian has existed for some time already and is continuing to develop at an incredible speed. Further, using ML to perform text analysis of the interview text files has promising and exciting opportunities to support an investigation of crime. For instance, ML can be used to find patterns and contexts related to places, names, organizations and events. The design of this project combines innovation and applied research, builds upon a “building-blocks”, experiences and expands the scope from an ongoing pre-study within applied research. The pre-study started in September 2020 as a cooperation between NTNU CCIS and Oslo Police District. We have so far learned that “speech-to-text” for Norwegian already exists in use in the market with inspiring results that have been found very promising by the Norwegian police. Thus, our research focuses on assessing the readiness of speech-to-text technology for police work, and also develops and assesses a user interface for such tools that the police would be ready to adopt.