Hum a few bars and I'll sing it for you...

Related link: http://www.musipedia.org/

Finding music online, or really in any large collection, has nearly always been limited to keyword searches. This means that the searcher is at the mercy of the person who writes the description in the track catalog. Sometimes we want something that sounds LIKE something else, or has a characteristic, or SOMETHING that can’t be captured in the keywords. You know – like how it sounds?

There are a few services out there now that demonstrate audio data-mining. Comparing wave forms and coming up with matches based on the actual sound, and not the tags or description. You get some interesting results this way. Looking up a sound at FindSounds www.FindSounds.com lets you search first on keywords, and then once you find something close, you can search on sonic similarity to that sound. This kind of data mining can become part of the creative process, as you find sounds that are similar that you would never have listened to based on a keyword search. For instance, I searched for Lion – found a reasonable Lion roar sound effect, and found a sonically similar file which turns out to be the sound of a pinball machine flipper triggering. And I can actually hear the similarity.

You can see how you could apply this kind of wave form datamining to music as well, although this kind of waveform analysis is not very musical. It would probably do a reasonable job of returning files that had some sonic resemblance – but what if you wanted a song that had a melodic resemblance? Or even better – have you ever thought “I know how it goes, but I don’t know what song it is…” I want to hum a few bars of the tune, and have the search engine tell me what song it is.

Well, now you can. Thanks to the Alexa Web Search Platform we have the wonder that is Musipedia. You can search through their growing library of songs based on MELODY as well as keywords and search phrases. The melody input is based on something called “Parsons Code” which is a very simple encoding of the melody which only records relative pitch movement of the notes. It completely ignores rhythm. There are only 3 symbols – ‘U’ for up, ‘D’ for Down, and ‘R’ for repeat. Everything is relative off the first note, which is notated with ‘*’. So, the first famous 8 notes of Beethoven’s Symphony No. 5 in Cmi, 1st Movement, 1st Theme is notated as ‘*RRDURRD’. This returns a surprising number of melodically similar pieces. I’ll leave their discovery as an exercise for the reader.

However, this isn’t the same as “Hum a few bars and I’ll sing it for you.” But don’t fret – at the top of the page you will see a link entitled “Search by Whistling”. Yes indeed, whistling, humming, or even playing a keyboard into your computer will work. Hum a few bars into your computer, the tune you hum is recorded, analyzed and a Parsons Code returned that you can then search with. This recorder/analyzer is a Java app, so it requires a policy file be put in your C:/Directories and Settings/Username/ directory. They give you instructions on the site, and even give you a link to the policy file that you can just download and save.

So while technically you CAN hum this, I found I got much better results by actually whistling – but that’s good enough for me as you can see in this slightly reduced screen capture of my efforts.

I can see this being applied to Sound Effects, where the mouth-made sound effect is captured and analyzed and sonically similar effects in the library are returned via a FindSounds type data mining interface. I mean, most sound guys can hardly talk without interjecting sound effects into their speech. It’s what makes us sound guys to begin with. Or maybe that’s just me.

This is still in its infancy, but I can see the Star Trek future where I say “Computer, find me a tune that sound like this… ‘la dee dum dee da’”, and Majel Barret’s voice will tell me just what that tune is, and give me options for many other tunes that sound similar. This is some very cool stuff.

Whistle while you work...