AI to detect danger in abused women’s voices

Can the voice of a victim of gender violence and the sounds that surround her be the key to her protection? Can artificial intelligence be useful in detecting sexist violence? Spanish researcher Esther Rituerto proved that he is and was rewarded for it.

Spain’s government delegation against gender-based violence recognized Rituert’s work in the latest edition of the Violence Against Women Doctoral Thesis Awards, for “deepening the use of audio technology and artificial intelligence to prevent and combat” sexist abuse.

Rituerto explains to EFE that his work was intended to be used speech and acoustic signals surrounding the woman to detect when the situation may be dangerous for her.

Namely, detect the fear in her voice in order to determine whether there is a danger of an aggressor attack, as well as the surrounding sounds that warn of danger: accelerated steps at dawn, blows, objects that break…

In its thesis, the Government Delegation for the fight against gender-based violence points out that artificial intelligence is used for understanding women’s reactions to risky situations or danger, with the aim of being able to generate automatic mechanisms for detecting these situations based on the auditory modality in particular.

Furthermore, he points out that “the entire methodological process used is extremely interesting and rigorous, and full of respect for the victims”.

The institution dependent on the Ministry of Equality believes that “the influence of this research in the medium term on the systems of protection of victims of gender violence can be very relevant”.

Discover the fear

Rituert’s thesis was born within UC3M4Safety, a project of the Carlos III University of Madrid that seeks to use technology to prevent, detect, protect and combat gender-based violence.

A research team was formed with profiles from different disciplines for the development of electronic devices capable of detecting fear in a victim of gender violence through voice or vital signs (temperature, pulse, sweating…).

And these reactions will prompt the smartphone to send an alert to emergency services and the police or the victim’s close circle.

The researcher, an audiovisual systems engineer, points out that the devices (a pendant with a microphone for recording sounds) and a bracelet (for detecting biological markers) are connected to a smartphone, which in turn is connected to a cloud system.

Artificial intelligence helps determine when the situation is risky and an automated response must be sent to assist the victim.

Rituerto specifies that these devices could be in operation in a few years (for now, tests are limited to the laboratory) and that in the near future they will begin to be tested on the street.

Victim status

The engineer was also involved in another research on the use of artificial intelligence to detect the status of a victim of gender-based violence through voice, that is, this technology is capable of determining whether a woman has suffered abuse by analyzing her voice. The developed system achieved an accuracy of 73%.

Rituerto emphasizes that an algorithm trained on victim data was able to spot tangible differences between women who were victims and those who were not, which related to the rhythm of the voice, intensity, number of words, speed of speech…

“We saw that he was able to separate victims and non-victims because there is something in his voice that distinguishes them,” he points out.

He currently works at the LMU Klinikum University Hospital and the Max Planck Institute for Psychiatry in Germany, where he researches the detection of mental illness using artificial intelligence in magnetic resonance images of the brain.