Three years and thousands of hours into the development of our music server, Octave, and we’re still not finished. (But we’re getting close).
It’s not the hardware that’s hard it’s the software, though perhaps not the software you might be imagining. Octave’s software has been able to pull and play music from any source for several years now. And sound amazing. No, the hard part is teaching a machine to know what you know.
You know that John, Paul, George, and what’s his name are the Beatles and not the Beetles. But who are Beatles, The? Or Fab Four? And how does Billy Preston relate to Beatles? And for that matter, who is Lady Madonna? Is she a singer (there is a Madonna who is a lady) or a song?
Teaching machines how to know what we know is really, really hard. And when a machine gets it wrong we think it’s stupid. But, in reality, it’s neither stupid nor smart.
The real challenges of building a world-class music management system are found in sorting through the millions of bits of data that have been collected on just about every recording made. Matching up the right track with the correct artist is just the beginning. There are no set conventions for how data are organized and no authority to guard over misspellings and inaccuracies as we might expect in something like the Oxford Dictionary (and even this classic isn’t perfect).
We are getting close. Over the past weeks, we have been testing the system and sussing out the quirks: assigning fuzziness to the logic where decisions aren’t black and white and scoring our accuracy against the standards of today (we’re at about 92% which is pretty darned great).
Octave’s getting close.