Samsung Event: Everything Announced Disney Plus Price Hike NFL Preseason Schedule Deals on Galaxy Z Fold 4 Best 65-Inch TV Origin PC Evo17-S Review Best Buy Anniversary Sale Monkeypox Myths
Want CNET to notify you of price drops and the latest stories?
No, thank you

Microsoft's online chief holds music search patent

Qi Lu, recently hired to oversee Microsoft's struggling search engine, received a patent in 2000 for a technique to search audio files and find similar songs.

When Microsoft hired Qi Lu to run its online business last week, the company trumpeted the fact that Lu holds 20 patents.

Patents are far from rare at Microsoft--many developers and researchers hold them--but the online business has typically been led by people with a business or marketing background. That hasn't been working out too well, so it's putting a geek in charge.

Qi Lu
Qi Lu Yahoo

The Seattle Post Intelligencer's Microsoft reporter, Joe Tartakoff, did a little digging on Tuesday to uncover exactly what kinds of patents Lu holds. Most interesting to me, one of them relates to music.

Specifically, it describes a PC application that could take a snippet of a song or audio file, break it down into component parts, analyze them, and then recommend similar songs.

It sounds superficially similar to what Shazam does, but the method is very different and more complicated. From what I can tell, Shazam simply takes a sound sample and matches it against a database with millions of audio files. Getting a fast result requires some fast data crunching, but there's not much deep analysis going on there.

Lu's patent (shared with two other engineers) proposed breaking the song all the way down to very small components like measures and individual notes, analyzing those components to find patterns--for example, a repeated sequence of notes might be the refrain or chorus--and then analyzing the relationships among those parts.

For instance, a pop song is typically constructed of several repeated verses and choruses, with a bridge somewhere in the middle. This is how the application would be able to identify and recommend songs that are similar to the song being played.

Instead of Shazam, the end result might have been more like Apple's recently introduced Genius feature, which builds playlists of songs based on the song you're currently playing.

I suspect that Apple's relying on data from all its iTunes users (Genius asks to collect data about your playing habits) and song meta data--for example, it often recommends songs by the same artist, or other artists in the same genre, or other songs released in the same era. That's much easier--both to program and for your CPU--than trying to analyze audio data for patterns.

Lu received this patent in 2000, which means that he was probably working on it several years before that. Check it out.