Way beyond Siri
Atkins and a small team of scientists in California and
Utah have developed software that recognizes words
and vocal qualities. The vocal data is run through algorithms
to infer, for instance, whether a counseling
session was empathic.

The first findings of automated evaluation of psychotherapy 
were published today in PLOS ONE. 

What does empathy sound like? It’s way beyond what Siri
could tell you on the iPhone. Think Hal 9000 from
“2001: A Space Odyssey.”

“Technology can provide a faster, less expensive feedback
mechanism to help counselors learn and retain their skills,”
Atkins said.

The current approach to evaluating budding psychiatrists and psychologists is decades-old:  Counseling sessions
are recorded on audio or video. People (called “coders” or “raters”) who have been trained to interpret these
sessions listen to them afterward and, as objectively as they can, analyze the therapist’s effectiveness.

 

The approach’s shortfalls are evident.

 

“It takes a ton of time, not only to find and train coders but
then they also have to watch or listen to a session and

analyze it in such a way that we can tell that Coder No. 1
and Coder No. 2, if they watch the exact same session,

would reach the same conclusion,” Atkins said. “There’s
lots of evidence that’s not always the case.”

The prospective solution (see model in PDF) capitalizes
on speech signal-processing models from the Signal

Analysis and Interpretation Lab at the University of
Southern California. 

“Advances in human behavioral signal-processing promise not only cost-savings by automating a process

typically done manually, they also enable new insights. The case of modeling empathy is a clear example,”

said Shrikanth Narayanan, the lab’s director and a USC professor of engineering.

Zac Imel, assistant professor of counseling psychology at the University of Utah, noted: “We have effective treatments,

but no practical way to evaluate counseling quality
in the real world. This technology offers one way to give
providers immediate feedback on what they are doing.”

The team’s ultimate goal: software that records and
generates real-time feedback during a therapy session.

For instance, a red light might indicate the therapist
is getting confrontational, or a blue light, tracking well
with the client.

“We’re trying to get at what’s happening in the session.
If we could tease out that information, we should be able

to identify with much greater specificity the things
counselors do that are effective.”

The research is funded by the National Institute on Drug Abuse and the National Institute on Alcohol Abuse

and Alcoholism – because the team is focusing on two psychotherapies that involve addiction treatment, Atkins added. 

Media contact for David Atkins: Brian Donohue - 206.543.7856; bdonohue@uw.edu
Media contact for Shri Narayanan: Amy Blumenthal - 917.710.1897; amyblume@usc.edu
Media contact for Zac Imel: Jana Cunningham - 801.581.3862; jana.cunningham@utah.edu