Move over, Freud? Startup says it can detect depression by analyzing your voice
Israeli firm VoiceSense claims its software can diagnose mental illnesses by listening to recordings of phone conversations
Shoshanna Solomon is The Times of Israel's Startups and Business reporter
Talking to a patient is one of the most important diagnostic tools doctors have to determine depression. Now, an Israeli startup claims it can help doctors diagnose depression and other mental illnesses through analysis of a patient’s voice.
“There are no biomarkers today to track depression,” said Yoav Degani, founder and CEO of the predictive speech analytics firm VoiceSense. “Our product gives an objective psychological analysis and is very accessible,” providing a picture not only of mental health but also of a person’s well-being, he said.
VoiceSense, founded in 2000, is undertaking a clinical trial of its speech-based predictive analytics technology for detecting depression at the Beer Yaakov Mental Health Center, the largest in Israel, to see if there is a correlation between speech patterns and depression and other mental illnesses, including schizophrenia. The trial will also see if it is possible to track changes in speech patterns, to understand changes in a person’s state of mental health.
The company collects a patient’s cellphone conversations with permission and analyzes not the content but rather prosodic, or non-content, elements of speech.
“The preliminary results are significant, and show typical speech patterns that characterize the depressive population vis a vis those that are not, and shows changes in speech patterns when a person is depressed,” said Degani in a phone interview.
The analysis studies a variety of parameters including the energy level of the voice, speech flow, its emotional level and sentiment, anxiety and stress, intonation, and pace and other indicators to capture the state of mind of the speaker, he explained.
“The software is language agnostic,” Degani said, and has already been validated in a large range of languages, including Western languages, from English to Polish and French, as well as Far East languages that are tonal — in which saying words with different tones changes the meaning of the word, even if the pronunciation is the same, such as Mandarin and Punjabi.
The company has been collecting speech pattern data and analyzing cellphone conversations for over a decade. It aims to provide a real time indication of a person’s frame of mind by comparing data gleaned from current conversations to a personality profile the software has created based on previous conversations.
More than 300 million people of all ages suffer from depression globally and the illness is the leading cause of disability worldwide, according to the World Health Organization. Some 6.7 percent of all American adults suffered from at least one major depressive episode in 2015, or some 16.1 million adults, aged 18 and older, according to the Anxiety and Depression Association of America.
“If we can predict the risk groups in advance, then people will be able to get better treatment and avoid hospitalization or prevent repeated hospitalization,” Degani said. Ideally, in future, people will be giving voice samples to doctors to monitor their well-being, just as today they give blood samples, he added.
Prof. Yechiel Levkovitz, director of the Beer Yaacov Mental Health Center and chairman of the department of psychiatry at the Tel Aviv University, said the preliminary results of the trial “are very good, and show the efficacy of the system.”
“When our patients go home, we pray we will be able to identify changes in their situation via their cellphones,” he said.
Others are also developing ways to diagnose illnesses through voice analysis. Psychiatrist and researcher Charles Marmar is collecting voice samples from veteran soldiers and analyzing cues like pitch, tone, rate, rhythm, for signs of post traumatic stress disorder, depression, and traumatic brain injury, while researchers at the University of South California have developed a tool called SimSensei that analyzes vowel sounds to detect depression and PTSD.
“There are many different approaches for leveraging speech-analysis in healthcare and wellness monitoring. Each has its own methods and focuses,” said Degani. “Our approach is highly accurate based on many years of research. And our technology is uniquely independent of language, culture and individual differences. Another comparative strength of our approach is that the technology works effectively in noisy environments.”
Other applications of the voice analysis technology developed by the firm address the commercial and enterprise market, providing enterprises, insurance companies and banks with analytics of conversations with customers to evaluate their risk of default or their loyalty to the brand.
The company’s technology has been able to predict if a customer who is calling the bank for a loan will be able to repay the loan or not, said Degani.
Isn’t this somewhat creepy? “All healthcare analysis is done by consent,” Degani said. “We cannot analyze a person’s voice without their consent.” And just like a person hands in a blood test for DNA testing, so people will have to consent for their voice to be analyzed.
Degani, who served in Israel’s elite 8200 intelligence army unit, working on signal processing, went on to become a clinical psychologist. The company, he said, “combines my two worlds.”
The firm has raised some $6 million to date from private investors and operates out of Herzliya.
VoiceSense is aiming to soon start a healthcare proof-of-concept phase of its technology in the US, Germany and other European countries, he said, where patients will be asked to use the technology in their day-to-day lives, as part of an experimental framework. The company is in parallel looking to enlarge its customer base for enterprise analytics, Degani said.