Components - Synchronisation

Human beings are almost unique in that they will entrain to an external stimulus or sound. What this means is they'll synchronise their bodily movements with it. Like dancing - they'll tap their feet or nod their heads in time with the rhythm of music.

In the natural world, humans have a preferred speed, or tempo, at which they like to walk. And that's a bit like music, which also has a tempo. But in the modern world, everything's changing, including music.

Live music is great, but there's no point having a band if you're not all playing to the same beat, which is why orchestras have a conductor.

For contemporary bands, keeping the rhythm together is important. Even bands that play live, such as the Gorillaz, might use samples or loops when they're playing. And they use something like a click track to keep them in time.

[CLICK TRACK CLICKING RHYTHM]

So the problem with the backing track is that it doesn't change, so the drummer's forced to listen to that to keep them in time. To try and solve this problem of getting a computer sequence to follow a drummer, the first thing we want to do is give it something to listen to. So we put up microphones on the drums. We've got a kick drum mic and a snare drum mic. And we put those through a process that's called onset detection. It simply says when he actually hits that drum - [DRUM HIT] - and that gives us event times of when they're hit. Then we have a click track from the computer of where it thinks the bar and the beat is, and the two have to be compared to give it a tempo output.

What's complex is that the drummer has a lot of techniques that you may not ordinarily think of here. It's things like syncopation. He's putting beats or humming in unusual places. And events can happen early or late. That's part of the feel of what a drummer does. A drummer isn't a metronome robotically playing a pattern. They're putting accents and moving the time around. So that really gives it something complex in terms of the rhythm. And what I've had to do here is try and create the best way for the computer to understand the information - almost to look around the beat, within a window, and use the timing information it's getting to make the best judgement it can of what the tempo is, and also not go wrong.

The glorious thing about live music is the ability to vary things according

to the audience and to your mood. So for example, you might want to slow things down a little bit - [DRUMMER PLAYING] - or speed things up. [DRUMMER PLAYING] Or you could get creative. [DRUMMER PLAYING] Easy now, Dave. [DRUMMER PLAYING]

Alan Turing was a famous mathematician and codebreaker. And in the '40s, he proposed an experiment that sets the standard for what it might be for a machine to be called intelligent. This is if a person cannot tell the difference in conversation whether they're talking to a machine or talking to a person. And here, we're going to apply the same principle, but to music. We're going to ask David if he can tell the

difference between me tapping and controlling a tempo and the computer controlling the tempo by listening to him.

I'm going to give you two options. I'm going to flip a coin. And if it's heads, then I'm going to tap and control the speed to match your drumming of the music that you hear. And if it's tails, then the computer system is going to control the speed. Both should feel like they respond to you, so feel free to try and experiment. Try and figure out which you think it might be, and then at the end of each piece, you guess which control you thought was in charge. Yeah?

Yep.

OK.

The less able Dave is to tell whether it's me or the software following his beat, the better the software is, and the more highly it scores on the Turing test.

I repeated the experiment

eight times - four with me doing the tapping, and four with the software doing the work.

There were, actually, through throwing this coin, there were four times that it was the tapper, me, and there's four times that the machine was controlling it. When it was the tapper, twice you guessed the tapper correctly, twice you guessed the machine. When it was the machine in control, you always guessed the tapper. In which case, I'm pretty happy with it. I'm happy with how it's performed, and that really, you can't tell the difference on that. And it's an interesting test.

Having music software that can respond to human musicians is a real breakthrough, and could revolutionise the future of music. Watch this space.

If you're interested to know more about the research or want to try out this software, feel free to contact me, Andrew Robertson, at Queen Mary University of London. You can also visit the website - that's b-keeper.org.

[MUSIC PLAYING]