What do you work on?

I study voice iden­ti­ty per­cep­tion, that is I look at how we recog­nise famil­iar voices, how we dis­crim­i­nate between unfa­mil­iar voices and also how we learn new voice identities.

“People must have indeed formed a rep­re­sen­ta­tion based abstract­ing voice-spe­cif­ic aver­ages as opposed to the trained exemplars.”

What did you do using Gorilla?

I have run quite a few studies on Gorilla by now. One of the first studies I ran was trying to work out how people form rep­re­sen­ta­tions of newly learned voice iden­ti­ties. When we hear voices in every­day life, one strik­ing feature of these voices is their within-person vari­abil­i­ty: the voice of the same person will sound dif­fer­ent depend­ing on each con­ver­sa­tion partner, speak­ing sit­u­a­tion and speak­ing envi­ron­ment (e.g. giving a job inter­view versus talking to a friend versus shout­ing at the pho­to­copi­er). Despite this within-person vari­abil­i­ty, we can recog­nise famil­iar voices with rea­son­able reli­a­bil­i­ty. So, how do we make sense of all of this vari­able and poten­tial­ly messy input? There are pro­pos­als in the face and voice iden­ti­ty per­cep­tion lit­er­a­ture sug­gest­ing that people form rep­re­sen­ta­tions when learn­ing new iden­ti­ties that are based on abstract­ing the aver­ages of all expe­ri­enced instances of that face or voice.

In order to test this pre­dic­tion, we decided to train lis­ten­ers to learn to recog­nise – and thus form a rep­re­sen­ta­tion of – 3 voice iden­ti­ties. The acoustic prop­er­ties of the train­ing stimuli were care­ful­ly manip­u­lat­ed such that each voice’s train­ing stimuli formed a dis­tri­b­u­tion that was missing its centre. So, in this set up, lis­ten­ers never heard the average acoustic prop­er­ties of the voices during training.

At test, we then pre­sent­ed lis­ten­ers with novel record­ings of the 3 trained iden­ti­ties that mapped onto the acoustic prop­er­ties they had heard during train­ing. Cru­cial­ly, we also now pre­sent­ed lis­ten­ers with record­ings that fell onto the average of the train­ing dis­tri­b­u­tions. If aver­ages are indeed abstract­ed during iden­ti­ty learn­ing, we would expect that lis­ten­ers are as good or even better at recog­nis­ing the trained iden­ti­ties from these average record­ings com­pared to the other recordings.

What did you find?

And this is exactly what we found: Lis­ten­ers were more accu­rate at recog­nis­ing the record­ings that cor­re­spond­ed to the unheard (!) average than the record­ings that fell onto the pre­vi­ous­ly heard train­ing dis­tri­b­u­tion. We could also show in a com­ple­men­tary analy­sis that accu­ra­cy in fact increased the closer a record­ing was to the average in terms of its acoustic prop­er­ties. We took these find­ings as an indi­ca­tion that people must have indeed formed a rep­re­sen­ta­tion based abstract­ing voice-spe­cif­ic aver­ages as opposed to the trained exemplars.

We are now looking into running follow-up exper­i­ments to find out more about the under­ly­ing mech­a­nisms that may be under­pin­ning these findings.

“It is amazing how quickly anyone can set up a study from scratch.”

Has this study been published?

Our article has been pub­lished by Nature Com­mu­ni­ca­tions.

How did Gorilla make your life or research better, easier or faster?

Switch­ing to online testing with Gorilla made a huge dif­fer­ence to me. It saved me time and hassle when setting up exper­i­ments and studies. There are also essen­tial­ly no com­pat­i­bil­i­ty issues when switch­ing com­put­ers or sharing studies. Addi­tion­al­ly, I sud­den­ly had access to par­tic­i­pants through­out the year, which had pre­vi­ous­ly been an issue.

What is the biggest advan­tage of online research methods?

By having my study online running in a browser, it now takes me one after­noon to collect my data as opposed to having to test in person for weeks and weeks.

How do you think online research is going to change your field?

It will be a lot easier to get ade­quate sample sizes when testing online, which improves the quality of research.

Did you include any special fea­tures in your study to ensure good quality data? If so, what did you do?

Our studies usually involve playing sounds to people, so we always need to make sure people can actu­al­ly hear our stimuli. There is a useful screen­ing (Woods, Siegel, Traer & McDer­mott, 2017) that we use at the start of our studies to make sure that people are wearing head­phones and are not lis­ten­ing to our record­ings on tinny laptop speak­ers. Through­out our tasks we also use audi­to­ry atten­tion checks to make sure people keep on paying atten­tion. These checks usually take the form of catch trials where instead of playing one of our exper­i­men­tal stimuli, we present a record­ing that is instruct­ing par­tic­i­pants to pick a certain response for this par­tic­u­lar trial (e.g. “Please press 7 now”).

For you, what is the stand-out feature in Gorilla?

It is really easy and intu­itive to use Gorilla. It is amazing how quickly anyone can set up a study from scratch. This is not only useful for my own studies but now stu­dents can easily create their own tasks for their research projects, which helps to really get to grips with their experiments.

Who or what orig­i­nal­ly inspired you to work in your field of research?

For my under­grad­u­ate, I studied lin­guis­tics and pho­net­ics, which set me off on the path of doing audi­to­ry research. How did I get inter­est­ed in this some­what niche field? Well – I watched My Fair Lady when I was a teenager…

When you’re not working, what do you enjoy doing?

Cycling, trying to find the deli­cious cin­na­mon buns or almond crois­sants in town.

What science book have you read recent­ly that you’d rec­om­mend to others?

Not nec­es­sar­i­ly a tra­di­tion­al (pop) science book but The Feather Thief by Kirk Wallace Johnson was an inter­est­ing read. I know a lot more about early nat­u­ral­ists, the exotic feather trade and fly tying after reading it!