July 29th, 2007
Singing search engines have it all wrong
On an off-topic item, I came across fellow blogger Roland Piquepaille’s blog about search engines that let you find music by singing to the computer. Well Midomi has just such a search engine where you sing to the computer to find the song you’re looking for. While Midomi sounds interesting, it flat out doesn’t work in practice. It’s not just the hardware requirements that are failing; it’s the human that’s the weak point.
One day I heard a song on the radio that I liked and I couldn’t catch the name of the song and I recalled hearing about Midomi from somewhere so I fired up the webpage as soon as I got home. I couldn’t really use this in the office because I didn’t have a microphone and even if I did I wouldn’t have used it because I would have felt embarrassed. I consider myself in the minority since I have a working Microphone hooked up since I have a Polycom Communicator but most people don’t so that’s one major limitation of these types of search engines. But even assuming that problem can be overcome, we’re still a long ways off from a working solution.
So when I fired up Midomi and got my Microphone working, I find my voice cracking because I hadn’t warmed it up yet and it’s been about 17 years since I’ve been in a Symphony Chorus so I’m out of practice. I finally manage to sing the right tune in to the computer but no luck finding the song because I only knew a few notes to the song. I ended up spending an hour typing in the few words of the song that I did know in to Google and I finally managed to find the song. So while the concept was certainly interesting, it was utterly useless from a usability standpoint.
So that got me thinking about how I would approach the problem in a way that combines the best of the text and note search techniques in an easier to use interface. While I’m certainly no pianist or anything close to one, I know I can hunt and peck out a few notes and I’ll bet most people can. If Midomi had something like this flash based piano for instance, I wouldn’t need a Mic hooked up and I wouldn’t need to warm up my voice. Even people who can’t carry a tune have a chance to hunt and peck the notes. Of course it wouldn’t need to be at the right key and the search engine could transpose through every key to search for the right song and you’d be able to make adjustments to the notes one at a time.
Furthermore, words could be directly attached to each note if you can visually see the notes which means the search parameters would be greatly narrowed. Even if all you had was five words correlating with six notes, that would almost precisely pin the song down. You don’t need a word for every note and there could be blanks left in place, but the more information there is the easier it is to narrow the search results. While there may be songs that share similar sequences of notes or similar sequences of words, the odds that they would share the same words corresponding to the same notes would be highly improbable.
Now granted, not everyone will be able to hunt and peck on a piano keyboard but every computer has a mouse while few have working Mics attached. The most logical solution would be to have both user interfaces available for the user and let them choose what they’re comfortable with or what they’re limited to. We’re not asking the user to play a whole symphony here; just a few notes. Each note can be wrong and it won’t be committed to the search parameter until the user hears the right note and confirms it. If you had a really smart search engine, real time possible results should start playing back as you’re pecking out the notes.
So to the people at Midomi or whoever else may be reading this, how about it? Can you give me this search engine that I’ve described?
George Ou is Technical Director of ZDNet. See his full profile and disclosure of his industry affiliations.
