by Jennifer O'Brien

Voice analysis to aid in fight against suicide and depression

Oct 11, 2018
Artificial Intelligence

UNSW associate professor Julien Epps is confident and hopeful the power of vocal analysis technology using machine learning will help in the detection and prevention of suicide and depression.

Epps is UNSW Sydney Faculty of Engineering’s Deputy Head of School (Education), and associate professor of the School of Electrical Engineering and Telecommunications. His specific expertise is in speech signal processing research.

“It’s a very interesting project. It’s an emerging area of research, which can hopefully have some very nice applications,” Epps told CIO Australia.

“Over the years, I’ve worked mostly on algorithms that can be used particularly to understand people’s emotion and mental state from speech. And so a key focus of that research then has been the detection of depression and pre-suicidality.”

Epps said there are two main components to the technology behind the project.

“One is speech signal processing. What comes into a system that we would develop is the speech waveform and then that’s analysed using signal processing methods – various sorts of methods for extracting features that are characteristic of people’s mental state.

“That is then passed to a machine learning backend. We use statistical machine learning to infer various properties about a speaker’s voice from that, based on labelled training data.”

Asked if this is a new area of research or builds on prior learnings, Epps said it’s a combination of the two.

“There have been various groups that have looked at this, particularly in the last five years or so. There was a little bit of more background scientific style research in earlier years, but I’d say the focus for engineering researchers has really picked up in the last five years.

“My group has been particularly active in this area. And particularly in either detecting depression from speech, or predicting the level of depression from speech.”

Asked about the far-reaching implications, Epps said the research is helpful to determine behaviour, but also opens the door to real-world practical use cases of the technology.

“It’s still something that is being understood. But there are a number of organisations very active in terms of use of technology for screening, intervention, helping people to services and so on.

“The role of this particular piece of research that I’m apart of is trying to detect the presence of depression or pre-suicidality, which could have a number of different functions. It could serve as a screening mechanism. It could be used as part of diagnosis – not full diagnosis – or it could be used for monitoring – if there’s some sort of treatment or follow up, for example. It might be that someone’s voice could be monitored over a period of time just to get a general indication whether things are improving or maybe going down hill.”

Helping crisis callers

At the same time, Epps is also using his expertise in vocal analysis using machine learning and applying it to a $1.1m five-year study of how Lifeline helps crisis callers. Today, about one million calls are made to Lifeline, the national telephone helpline.

As part of the study, Epps and his team will determine how machine learning can be used to analyse the vocal tone of people calling in to identify the kind of help they need. “My team will lead the use of artificial intelligence methods to automatically identify different types of help-seekers based on their vocal qualities,” Epps said in a UNSW media statement, explaining AI will also examine written communication in cases of online chat and SMS text. “The analysis of acoustic and linguistic information from the speech of crisis callers is a highly novel research area, where there is significant potential for new technology to contribute.”

He expects machine learning couldbe ever-present, listening in to calls to triage them while contributing to Lifeline’s long-term strategies for supporting help-seekers.

“The crisis supporter will probably always be the first responder, but the use of machine learning could be a cue to get other specialists involved part-way through the call, or for example, help assess how many distressing calls a crisis supporter has had to handle,” he said.

Academics from eight Australian universities and one in the US will be contributing to the study, with backgrounds in psychiatry, psychology, sociology and engineering.

Chief Investigator leading the study is University of Canberra’s Professor Debra Rickwood, who developed the idea after doing some evaluation work with staff at Lifeline.

“Lifeline has moved into the digital age and offers crisis support via online chat and soon via SMS text messaging,” Professor Rickwood said.

“Yet, despite increasingly widespread reliance on Lifeline, little research has identified the types of help-seekers that such crisis services are expected to support, nor the outcomes expected to be achieved.”

For Epps, he told CIO Australia he wants to be able to make a “helpful contribution” in the area of depression and suicide prevention and detection – and hopes this study, along with his other projects, will help him achieve that goal.

“I haven’t historically worked on mental health at all, and I’m still fairly new to the area, but I think it’s very exciting that technology could play a role here. It always astonishes me to learn how many people are touched by some kind of mental health problem and difficulties.

“If technology can play some part in helping those people, then I think that’s tremendously motivating.”