Analysis of millions of audio files has led one US company to claim that their software can predict how a person’s voice will make a listener feel
Thursday 16 April 2015 07.56 EDTLast modified on Friday 17 April 2015
Few people consciously think about their voice, but the way we speak is one of the most fundamental parts of our individual identity. The intricate acoustic patterns which comprise speech affect how we’re seen in terms of our personality, our emotional state and even our professional competence, but it’s only been relatively recently that scientists have tried to delve into these complex vocal traits in more detail.
The first people in recorded history to study the human voice were the ancients Greeks, a race fascinated by the differences in the vocal spectrum and their varying impact on our emotions. Sometime in the 2nd century AD, the Athenian scholar Julius Pollux attempted to meticulously map out the full range of human vocal qualities as perceived by the listener.
Little survives of his work, apart from a detailed list describing how voices can range from being engaging or feeble to persuasive and even melodious. Pollux noted that the latter gave the impression of a cultivated individual.
This is typical of male voices which the majority of listeners found extremely engaging when responding to the question, “Greet me as if I am a new customer”But what actually makes a voice seem engaging or melodious? Two millennia later, the decidedly 21st century combination of big data and crowdsourcing may be yielding the answer. Based on the analysis of several million audio files and input from thousands of recruiters and US consumers, the company Jobaline has created a technology which can predict how a person’s voice will make the listener feel. Studies performed by independent market research firms have found it has an accuracy of around 75%.
The technology is aimed at streamlining recruitment in the hourly jobs market, a source of employment for over 76 million people in the US. It’s found a niche as for certain industries, having the right voice is a vital requirement for the job. “Hospital call centre services often require voices which will make people feel calm when handling delicate matters,” Jobaline CEO Luis Salazar explains. “Retailers want voices that Generation X and Generation Y will find appealing to engage with both over the phone and at the checkout. Surveys have shown banks prefer to hire people with soothing or calming voices.”
Psychology has shown that certain cues in speech, known as paralinguistic elements, are more important than others when it comes to generating particular emotions in the listener. Low frequency, continuous sounds are far more comforting to distressed individuals than intermittent sounds, acoustic techniques which mothers tend to naturally use when trying to soothe newborn babies.
This female voice is typical of those which the majority of listeners found extremely calming when responding to the question, “Can you describe what customer service means to you?”In the absence of face-to-face interaction, we also tend to pick on a couple of traits to determine whether a voice sounds trustworthy or not. Intonation, meaning the pattern of the pitch changes in a voice, is very important with male voices that are low pitched and follow a rising melodic trend being deemed particularly untrustworthy. The same is true for low pitched female voices which tend to fall while speaking.
Specific patterns in our speech also go some way to determining how people perceive us both socially and at work. “Speaking rate seems to be one of the most reliable attributes when it comes to how people judge a person,” says Jennifer Pardo, who researches speech communication and phonetics at Montclair State University. “We pick up on features like how long the segments are when they’re speaking. In general people who speak a little slower tend to be perceived as more friendly or benevolent while we often associate things like competence and authority with people who speak a little faster. But there’s a certain sweet spot to it, if you speak too fast then you sound nervous.”
It’s not really clear how we evolved to develop these intuitive preferences but they can also vary somewhat from one person to the next. “We might think that everyone sees a high pitched voice as being happy but that’s not the case,” Pardo says. “If you’re looking for one particular acoustic element which reliably determines how the majority of people feel about a voice, the closest is speaking rate, followed by intonation but it’s not one thing on its own.”
The variation involved means that accurately predicting which voices will be best for a particular job, independent of age, gender and ethnicity, means pinpointing the right combinations of vocal features. Jobaline’s approach was to identify interactions between an array of different features from pitch to energy accumulated over time, which all combine to give each voice its unique fingerprint or signature.
Leonardo DiCaprio in The Wolf of Wall Street. Despite the fact that his tone is aggressive in this clip, the underlying essence of his voice remains engaging.“Everyone has a core voice signature made up of a series of things,” Salazar says. “To see what I mean, listen to Leonardo DiCaprio screaming with passion in The Wolf of Wall Street and then listen to him in one of the romantic scenes in Titanic when he was way younger. The underlying essence of the voice is the same but when we look at the energy accumulated on a specific frequency domain and combine that with other acoustic elements, this voice scores above the threshold that the average listener in the US would consider an engaging voice.”
While each voice signature is unique, it’s also possible to group collections of voices based on the emotions they elicit. Jobaline exposed thousands of Americans to these groups and asked them to vote on how they felt and their software is based on the fact that the majority of people found certain groups more calming while other groups were more engaging. But this may not apply globally, as our preferences are also heavily culturally related. While 75% of Americans may find a certain voice highly soothing, British or Australian listeners may disagree.
But with this understanding of what makes a voice inherently suitable for a particular situation, is it possible to teach anyone to speak in a more calming fashion? “It’s not quite that straightforward,” Pardo says. “While you can train someone to say speak a little more slowly in a stressful situation, it’s also about vocal dexterity, the way a person uses their voice and certain people are more flexible with that. It comes down to motor control. It’s similar to the way a person walks. It’s not just about the size of their limbs, it depends on how well they can move them. A dancer for example, will move differently to a non-dancer.”
Leonardo DiCaprio in Titanic. His voice scores above the threshold that the average listener in the US would consider to be engaging.Having the right voice signature for a situation largely comes down to the extent to which you can control different aspects of your voice. Certain people can exert greater motor control over their voice than others. Charismatic politicians and CEOs are good examples, with the ability to consciously or subconsciously push their voices through a wide range of frequencies when addressing the general public to project an image of power and dominance.
Salazar believes that the ultimate goal is to make such powers of communication accessible to everyone and help create better job-worker matches.
“Beyond the application in the recruitment space, we’re hoping to work with research teams across the US to eventually use the technology to create computerized voices which generate an emotional connection with listeners,” he says. “This has applications in the field of Augmentative and Alternative Communication. Will it be possible to help people with a disability which forces them to speak through a device, to express the emotion they want to express?”
Salazar is optimistic, but the challenge will be as much about understanding the listener as manipulating aspects of a voice.
“The talker’s voice is just one part of the puzzle,” Pardo says. “The key thing is, given this particular listener, how should the talker behave? For example, when a particular person speaks fairly slowly, I might not like that as much as somebody else does. So you would need to be able to work out that sweet spot and adjust it for every person.