A research team at Universidad Carlos III de Madrid (UC3M) has developed technology using advanced machine learning techniques that detects signs of gender violence from paralinguistic characteristics of the voice such as tone, rhythm, and intensity. This innovative method helps to recognize situations of psychological stress or trauma while preserving the privacy of the speakers, which could have major implications for telephone helplines and telemedicine services.

The research, recently published in the scientific journal Applied Sciences, has developed a technology that works using an architecture called adversarial, which allows the recognition of people who have been victims of gender-based violence based on biomarkers related to spectral aspects of the voice. "This type of interpretation of speech characteristics is very similar to what we humans do intuitively. What our study does is transfer that knowledge to neural networks that, to a certain extent, mimic how the human brain processes this type of information," explains one of the authors of the study, Carmen Peláez Moreno, professor in the Department of Signal Theory and Communications at UC3M and researcher at UC3M4Safety.

To conduct the research, the team worked with volunteers who participated in experiments designed using virtual reality. During the tests, participants watched videos with and without violent content, while changes in their behavior and voice were analyzed based on the emotions they experienced. “From these recordings, we observed that there were very different behaviors in response to the same stimuli between people who had suffered violence and those who had not,” says the researcher. “It was a serendipitous finding: while looking for something else, we discovered that it was possible to detect whether a person had been a victim of violence simply by analyzing their voice.”

This breakthrough opens the door to important practical applications. On the one hand, the technology could be used as a support tool for the early and non-invasive detection of mental health problems in clinical settings. On the other hand, it could be integrated into digital platforms such as virtual assistants or social care resources. This would facilitate the early identification of victims of gender-based violence and help reduce the problem of underreporting, offering specialized support more quickly and effectively.

“If we can identify signs of gender-based violence when a person calls a helpline, goes to the doctor or a social service, we can act before a fatal event occurs, and even before the person themselves recognize that they are a victim, which would facilitate their psychological recovery, which must begin long before cases reach the media,” concludes Carmen Peláez.

The research is part of the Bindi project, developed by the UC3M4Safety team, which aims to combat gender-based violence by preventing assaults, collecting evidence, and providing early assistance to victims through technology. The UC3M4Safety team is led by Celia López Ongil and Clara Sainz de Baranda Andújar, and includes research staff from the Institute for Gender Studies (IEG), the School of Engineering, and all UC3M faculties, bringing together specialists from more than fifteen areas of knowledge, including engineering, social sciences, and humanities.

“The team has sought to use technology to solve social problems because we believe it can greatly help in the fight against violence and victimization, as well as in helping victims recover from their situation,” concludes IEG director Celia López Ongil, professor in the Department of Electronic Technology at UC3M.