Now Algorithms Are Deciding Whom To Hire, Based On Voice | KERA News

Now Algorithms Are Deciding Whom To Hire, Based On Voice

Mar 23, 2015
Originally published on May 27, 2015 7:58 am

If you're trying out for a job in sales, the person who judges your pitch may not be a person — it could be a computer.

Job recruitment is the newest frontier in automated labor, where algorithms are choosing who's the right fit to sell fast food or handle angry cable customers, by sizing up the human candidates' voices.

Let's take a voice you know: Al Pacino. Think back to how he sounds in The Godfather, Devil's Advocate, Scarface or this recent interview on Charlie Rose.

The actor speaks with different accents, different emotions, different ages — and his range is stunning. But in every version, Pacino's voice has a biological, inescapable fact.

"His tone of voice generates engagement, emotional engagement with audiences," says Luis Salazar, CEO of Jobaline. "It doesn't matter if you're screaming or not. That voice is engaging for the average American."

Years and years of scientific studies and focus groups have dissected the human voice and categorized the key emotions of the person speaking.

Jobaline has taken that research and fed it into algorithms that interpret how a voice makes others feel and then cross-checks its judgment with real human listeners. It's a departure from other data science. With facial recognition, for example, algorithms sift through your smile, your brow, to decide your mood.

"We're not analyzing how the speaker feels," Salazar says. "That's irrelevant."

Regardless of whether you're happy, sad or cracking jokes, your voice has a hidden, complicated architecture with an intrinsic signature — much like a fingerprint. And through trial and error, the algorithms can get better at predicting how things like energy and fundamental frequency impact others — be they people watching a movie, or cancer patients calling a help line.

Through machine learning and multiple feedback loops, it keeps answering and homing in on Salazar's question: "What is the emotion that that voice is going to generate on the listener?"

So far, Salazar says, the Jobaline secret formula can pinpoint if a voice is engaging, calming, and/or trustworthy.

Note: It's not a lie detector test. You could be a big liar, but just sound like someone honest.

Use It For Hiring

Big companies pay Jobaline to help them sift through thousands of applications to find the right workers for their hourly jobs. Human recruiters make the final judgment, but the startup determines the small pool that gets human consideration.

Jobaline says it has processed over half a million voices for positions including sales, janitorial staff and call center workers.

"In the hospitality industry, in the retail industry, you want people engaged. The average span of attention is four seconds," Salazar says.

That's very short.

The benefit of computer automation isn't just efficiency or cutting costs. Humans evaluating job candidates can get tired by the time applicant No. 25 comes through the door. Those doing the hiring can discriminate. But algorithms have stamina, and they do not factor in things like age, race, gender or sexual orientation. "That's the beauty of math," Salazar says. "It's blind."

Career Counseling

As a woman who has built a career on talking, I'm curious what the algorithms have to say about me. My friends say I've got two voices: the inviting, empathetic "Hey how you doing, come on over" voice. And the "Don't mess with me. I'm getting work done" voice.

Salazar ventures to guess the intrinsic quality: "I'll say it's engaging and trustworthy. I don't think it will make the bar for calming. We'll see."

The algorithms agree. They say, with 95 percent certainty, that my voice is engaging to three-quarters of Americans.

So, I'm a good fit for radio.

Copyright 2017 NPR. To see more, visit http://www.npr.org/.

ROBERT SIEGEL, HOST:

And now, All Tech Considered.

(SOUNDBITE OF MUSIC)

SIEGEL: If you apply for certain jobs, the way you present yourself - how you sound - may be considered. And soon in some industries, computers may be making that consideration. Job recruitment is the newest frontier where algorithms are choosing who's the right fit to sell food or handle angry cable customers. NPR's Aarti Shahani reports.

AARTI SHAHANI, BYLINE: Let's take a voice you know and play a few samples of it. Clip number one...

(SOUNDBITE OF FILM, "THE GODFATHER")

AL PACINO: (As Don Michael Corleone) My offer is this - nothing. Not even the fee for the gaming license.

SHAHANI: Clip two...

(SOUNDBITE OF FILM, "DEVIL'S ADVOCATE")

PACINO: (As John Milton) He gives you this extraordinary gift and then what does he do? I swear, for his own amusement...

SHAHANI: And number three...

(SOUNDBITE OF FILM, "SCARFACE")

PACINO: (As Tony Montana) I'd kill a communist for fun, but for a green card, I'm going to carve him up real nice.

SHAHANI: This is Al Pacino in "The Godfather," in "Devil's Advocate" and in "Scarface" - three different characters, three different accents at different ages. The movies were years apart. But in every version, Pacino's voice has a biological, inescapable fact.

LUIS SALAZAR: His tone of voice is - generates engagement, emotional engagement with the audiences.

SHAHANI: Luis Salazar is CEO of Jobaline.

SALAZAR: It doesn't matter if you're screaming or not. That voice is engaging for the average American.

SHAHANI: Years and years of scientific studies and focus groups have dissected the human voice and categorized the key of emotions of the person speaking. Jobaline has taken that research and fed it into algorithms that interpret how a voice makes others feel and cross-checks its judgment with real human listeners.

SALAZAR: We're not analyzing how the speaker feels. That's irrelevant.

SHAHANI: Regardless of whether you're happy or sad, cracking jokes, your voice has a hidden, complicated architecture with that intrinsic signature, much like a fingerprint. And through trial and error, the algorithms can get better at predicting how things like energy and fundamental frequency impact others, be they people watching a movie or cancer patients calling a help line.

SALAZAR: What is the emotion that that voice is going to generate on the listener?

SHAHANI: So far, Salazar says, the Jobaline secret formula can pinpoint if a voice is engaging, calming and/or trustworthy. And note - it's not a lie detector test. You could be a big liar, but just sound like someone honest. Salazar plays me a clip of a woman applying for a job.

(SOUNDBITE OF ARCHIVED RECORDING)

UNIDENTIFIED WOMAN: So one person that motivates me every day is my son because I'm trying to make a better life for him.

SALAZAR: So this is an answer to this - question what motivates you?

SHAHANI: Big companies pay Jobaline to help them sift through thousands of applications to find the right workers for their hourly jobs. The startup says it's processed over half a million voices for hotel receptionists, call-center staff.

SALAZAR: In the hospitality industry, in the retail industry, you want people engaged. The average span of attention these days is four seconds. In four...

SHAHANI: I'm sorry, can you repeat that?

SALAZAR: (Laughter).

SHAHANI: And the benefit isn't just efficiency, cutting costs. We humans can get tired by the time applicant number 25 comes through the door. We can discriminate. But algorithms have stamina, and they do not factor in things like age, race, gender, sexual orientation.

SALAZAR: Correct. Math is blind, basically, right? That's the beauty of math - it's blind.

SHAHANI: Now, of course, as a woman who's built a career on talking, I'm curious what the algorithms have to say about me. My friends say I've got two voices - the inviting, empathetic hey, how-you-doing, come-on-over voice, and the don't-mess-with-me, I'm-getting-work-done voice. Salazar ventures to guess the intrinsic quality.

SALAZAR: I'll say it's engaging and trustworthy. I don't think it will make the bar for calming, (laughter) but we'll see.

SHAHANI: The algorithms agree. They say with 95 percent certainty that my voice is engaging to three-quarters of Americans, so I'm a good fit for radio.

AUDIE CORNISH, HOST:

That was NPR's Aarti Shahani, who was hired by humans, not algorithms. And we'd like to see what the algorithms have to say about you. NPR's tech team is collecting vocal samples. Go to npr.org/alltech to submit your voice clip. Transcript provided by NPR, Copyright NPR.