Cowell v computer: Have talent show judges had their day?
- 19 April 2012
- From the section Technology
Music talent show judges are among the highest-profile figures in the modern media, watched by millions each week on shows like The Voice and X Factor.
But they can be expensive and temperamental. What if they could be replaced by computer judges - cheap, consistent and low maintenance?
This is essentially the task that Dr Nick Collins from the University of Sussex has set himself, with his research on machines which listen and learn.
He has programmed three computerised judges, complete with general musical knowledge and individual quirks, for a competition in London.
Dr Collins has achieved his feat thanks to a highly versatile programming language written specially for music, called SuperCollider.
Its devotees - from cutting-edge musicians to scientists and sound artists - have come from all over the world to London this week for a conference to showcase this powerful tool, which programmers can use to explore musical artificial intelligence.
But can a computer really have an "ear" for music? And if so, are the days of celebrity judges like Louis Walsh and Amanda Holden numbered?
Learning the rules
Type in that simple line of code and you will get a continuous, purely pitched sound - an annoying buzz.
Expand that into several thousand lines of code, and you could have a symphony of electronic sound.
Or better still, as SuperCollider is a dynamic, interactive programming language, you could teach the software to devise its own album of masterpieces - in any genre, from Peruvian pan pipes to acid jazz - while you go and put your feet up.
As a language, it is capable of a form of artificial intelligence. To create its own music, it can be programmed to follow set rules, known as algorithmic composition. Or it can be "fed" a diet of music - for example, exclusively reggae - in order for it to "learn" the rules of that genre, and in turn create its own songs.
This diet method is how Dr Collins created his judges, which were made for a Dubstep competition.
"The judge is a machine listening system," explains Dr Collins.
To be trained up as Dubstep judges, they need to be exposed to the genre.
The judges "listen" to several hours of early Dubstep, recent commercial Dubstep and general electronic dance music, which provides them with a general grounding in those genres.
The computers are analysing complex information within each track, the pitch, the timbre or "texture" of the music, the rhythm, the way the tracks change over time, and building an impression of what Dr Collins calls their "characteristic sound world".
"My judges also have pet hates and guilty pleasures," says Dr Collins. This is achieved by allowing each judge to analyse additional examples of dubstep, which will help shape their individual preferences.
It takes eight hours to train the judges. "I can't anticipate how they will turn out," says Dr Collins. "They are definitely independent entities now, and have their own biases.
"The judges' listening capacity is not yet on a par with a human ear, but this is an attempt to respect some of the ways that human hearing operates."
"They're never going to be as annoying as Simon Cowell," he says.
They also lack physical form at the moment but Dr Collins anticipates turning his creations into robots, so you can see them deliberating over their choices.
Generating music as a computer, as opposed to a traditional physical instrument, brings with it many advantages.
An instrument relies on consistent inputs: the pluck of a string, hard breath into a reed.
SuperCollider can take any kind of varying information, and turn it into a composition. Cues can include anything from the weather to a torch being shone on to a web camera - a trait that sound artists have made use of.
Or, it can learn to play along live with a musician, making it a kind of sensitive session musician.
In the same way that Keith Richards might know to play a certain kind of riff when he recognises Mick Jagger make a particular vocal rasp, SuperCollider can learn to respond to an individual artist. It can trigger an effect like reverb or distortion, or a drumbeat, on hearing a given sound, so it can take part in a live performance.
This ability is being explored in another SuperCollider event during this year's conference.
Coding by inspiration
Groups of coders have been assigned to collaborate with three very different musicians: a former world beatbox champion, a progressive metal guitarist and a flautist.
The teams have been given two days to rehearse together, before taking part in a live show.
Beatboxer Bellatrix has been assigned SuperCollider enthusiasts Chris Kiefer from Brighton, Yorgos Diapoulis from Greece, and Tim Walters, an Apple engineer who has come over from the US.
Rehearsing in the music department at Goldsmiths College in London, the coders sit armed with their laptops, watching waveforms that describe the kind of noise that the beatboxer is making.
They message each other lines of code as inspiration comes to them. Each line, or algorithm, will twist her signature sound in new directions.
"The algorithms listen to what she is doing with her voice and create a response, to create a system she can improvise with," explains programmer Chris Kiefer.
"We're trying out lots of algorithms to see what works with her voice."
Once they have settled on their "soundscape", they will be ready for a live performance in front of an audience.
"There are elements that are really comparable to working with other humans," says Bellatrix.
"The computers listen to what you say and respond - sometimes you can vibe with it, sometimes you say, that is mental! It's pushing me as a musician."
Jukebox jury out?
While artists like Bellatrix may relish the creative opportunities of software like SuperCollider, others are more sceptical about its ability to judge music.
"As a DJ, I still feel there is a decision I make to play or not to play a track based on very small subtleties," says BBC 6 Music DJ Nemone.
"One track can be nearly right, but not quite, so I don't play it, and very similar to something else I love, and will definitely play."
"I would say that feeling, that final emotional decision, can't ever be digitised."
An emotional decision is also made by DJs at live events, she adds, when they read the dancefloor.
"People listen to your choice of tunes and have come to expect and trust you'll play them a certain kind of track. I think you'd notice significantly if a computer was making that decision."
Until the emotional intelligence of computer software is cracked, perhaps TV's talent show judges are safe in their seats.