Gallery

Making the cut

Posted on 30 Mar 2009 at 16:44

Jon Honeyball discusses the future of Microsoft following its revenue upset and reveals his concert pianist roots.

Plug in the values for a 1080p screen at 42in and you get a value of 1.65m for the viewing distance, and it gets more interesting still once you get to PC screens: the 30in Apple monitor I'm writing this on has 2,560 x 1,600 pixels, which yields a viewing distance of 0.8m, and that's right about where I'm sitting. Now let's take a look at a couple of modern laptops. The Dell Studio 15 has a 15.4in screen with a resolution of 1,280 x 800. Plug those figures in and you get a viewing distance of around 0.8m, which means you're definitely going to be seeing those pixels at a normal distance. But what about a netbook with an 8.9in screen running at 1,024 x 600? Now you're down to 0.65m.

This spreadsheet forces you to think about resolution, size and distance, and the relationship between them. For example, there's no point buying a 42in 1080p screen if you're planning to sit 4m away from it. And give some thought to the size, resolution and viewing distance of the screen you sit at for hours: sit too far from it and you'll bring on eyestrain; sit too close and you can see the jaggies; get the right balance and it will help.

The singer, not the songsmith

In the past I've been very critical of the R&D spend at Microsoft Research, because far too little of those billions appear to have rubbed off on real product improvements. Much of what I've seen there either in person or via the MSR website has merely prompted a disappointed "uh-huh". Some stuff has been brilliant - the Paint touch-up tool that managed to work out how to remove someone from a photograph was brilliant - but such flashes of brilliance have been tempered by a sense of ennui.

Nevertheless, as someone who studied for many years to be a concert pianist, I'm always intrigued by the possibilities of using "artificial intelligence" to enable the musically inept to create something new. Now along comes Microsoft Songsmith, which encourages you to sing something into it and then works out what sort of thing you were attempting. It takes a good stab at extracting tempo, rhythm, timing, key and so forth from your burblings and then comes up with backing arrangement for you to sing along to.

Naturally, it works best if given something in 4/4 time and a relatively straightforward 16-bar verse structure. This is very clever software that will no doubt convince millions of gurning pre-teens that they're destined to be the next big thing.

Now compare Songsmith with the new release of GarageBand from Apple - which didn't consume trillions of Microsoft Research dollars in coming up with a self-learning backing track that attempts to do the impossible. Instead, it features a large collection of tuition videos, fully interactive, designed to teach you how to play the instrument in question - whether piano, guitar or so on. I was immediately reminded of that old proverb about giving a man a fish versus teaching him how to fish for himself. One of these products reduces all music to AI gloop, while the other teaches you to express yourself by playing. Could the difference be any more profound?