Search Music by humming, not perfect, but feasible
Web August 13th, 2008
I once asked my colleague the name of a song, I barely remembered its lyric, though managed to hum the melody, and I am pretty sure it come from his favorite singer, Fish Leong. But neither of us could recall the name of the song. I searched the web, and found Medomi:
Midomi stands out from other music search engines by supporting to search music by humming. The user may use its microphone to upload her humming, then the search engine would return the matched section. A full version purchase link(via iTune) sits besides it for the user’s convenience.
I hummed the song, and the very first return result is exactly what I am looking for. No a big surprise as the song is just released in Leong’s last album, and the artist is quite popular. Then I tried some old songs, more precisely, 1995 by Bob Chen. Nothing relevant returned, it seems the song not even in their database. So I registered, logged in and recorded my own rendition in Midomi studio, the next time I searched it by humming, the search engine returns the expected result.
As we all know, search, index are formalized, the hardest part is how to extract the pattern precisely and concisely. Midomi addresses this problem with an extremely brilliant idea, let the users do it, is there any delicate artificial intelligence algorithm smarter than human being? And the user activities are easily synergized for other web 2.0 ingredients, like friends, groups etc, that is a good news for the venture capitals.
The only missing piece is there is no API for the developers, that make it hard to integrate this web service to your personal music management software, like iTune or Amarok. As the pattern extraction is powered by human, if you happen to be in the tip of the long tail, just pray the singer get talented.
The other problem is, what if the user is tone-def?
Wonder how it would work for songs back from the 1940’s or 50’s? Many of those have probably not been catalogued.
Most likely it won’t work very well if either the user is tone-deaf or the targeted songs have not been hummed by other users.
The site tries to stimulate the users’ contribution by popularity contest inside users and their beloved artists. It works, at least in my first visit. I could not find my favorite songs from Bob Chen, so I hummed them to help whoever share the same interest.
It doesn’t work.
At first I hummed some songs and it wasn’t able to recognize them. “OK”, I thought, “maybe they are’t in the database”. So I tryed again with some song that were there for sure… with the same results: nothing.
Eventually I played a CD and I put the microphone close to the speacker: the song was aquired pretty well, but Midomi still failed. It was a complete disappointment.
I am not quite familiar with the algorithm which powers Medomi, the wild guess is the front-end extracts the melody from humming and matches the pattern in the database. If that is the case, humming is far easier to handle, no background chorus, no subtle drumbeats.
It is merely a toy.
Midomi stands out from other music search engines by supporting to search music by humming.
“I Agree with this Statement”
I really do not know how this works. I do not even know the basics of music. But i am just a bathroom singer. According to me, it will take a note of pitches and match with the database.
what a great post, i like it
Thanks for this Great Information :)
Very usefull information to all of us, especially i like your blog design, Very Attractive Design & Try to Add some Social Media Widget This will increase your blog visibility.
I will bookmark this url to my facebook Account, will visit again.
Britney
you guys rock!
“The only missing piece is there is no API for the developers”
Do anyone knows if it is a search engines API that alows us to search by recording a song?
Thanks!
Erik