AI speech analysis optimises suicide prevention hotlines

An experimental AI model could help suicide prevention workers respond to calls more effectively

Spotted: According to the World Health Organization (WHO), more than 700,000 people take their own life every year. While suicide hotline counsellors work hard to provide vital assistance, they are not always able to properly assess the severity of a caller’s condition.

Concordia University PhD candidate Alaa Nfissi has developed a way to use AI speech analysis to help. Nfissi has developed a model for speech emotion recognition (SER) using artificial intelligence tools. The model analyses waveform modulations in callers’ voices to identify people who are at highest risk of carrying through on committing suicide.

Nfissi used a database of actual calls made to suicide hotlines, alongside a database of recordings from actors, to create the model. Both sets of recordings were annotated to reflect a specific state of mind: angry, neutral, sad, or fearful/concerned/worried.

The deep learning model then analysed the data using a method that can detect and identify emotional states at a certain time. In initial tests, the model correctly identified fearful/concerned/worried states 82 per cent of the time. For neutral, sad, and angry states, the success rates were 78 per cent, 77 per cent, and 72 per cent respectively.

Nfissi’s findings were published in a paper, ‘Unlocking the Emotional States of High-Risk Suicide Callers through Speech Analysis’, on IEEE Xplore.

Helping those who suffer from mental health issues is the subject of a number of recent innovations, including an app that helps ease eco-anxiety and a platform that screens children for mental health issues.

Written By: Lisa Magloff