Join 3,415 readers in helping fund MetaFilter (Hide)


Nifty Audio Projects
April 2, 2011 6:43 PM   Subscribe

Nifty audio projects from Paris Smaragdis, including fascinating method of extracting individual audio samples (say a guitar solo) from a mix by humming the part. [6.4 mb mp4] [via AskMe]
posted by odinsdream (12 comments total) 40 users marked this as a favorite

 
I found this through the Askme today and thought it was wonderful. SOme of the waveforms were obviously munged after the sound extraction, but the applications for this are way cool.

You can get uncle Morty to cough again the makes sure you're able to remove it from your wedding video. You can unflatten a flat sound file so you can apply reverb to just the vocals. On and on. Neat stuff.
posted by cjorgensen at 7:05 PM on April 2, 2011


This looks really interesting. Thanks, odinsdream.
posted by flapjax at midnite at 7:08 PM on April 2, 2011


Very cool stuff. Can anyone find the technical papers describing his work? The link on Paris' site seems to be broken.

Methods like this are very impressive, but they always leave behind a lot of artifacts. It makes me wonder if someone could achieve better results by opening a spectrogram in Photoshop and using the clone tool to clean stuff up. Though you'd have to be careful about getting the phase data right.
posted by MrFTBN at 7:18 PM on April 2, 2011


D'oh. I found them.
posted by MrFTBN at 7:26 PM on April 2, 2011


This is really fantastic. I've wondered for years if there was a way to compare two audio tracks (say, a song with full vocals and an instrumental mix of the same song) and extract the difference (in this case, to isolate the vocals). But this tool seems to go even farther. Very impressive.

I can't wait to hear the mashups that result from this.
posted by The Winsome Parker Lewis at 7:33 PM on April 2, 2011


I've been trying to read the paper "User Guided Audio Selection from Complex Sound Mixtures" but it gets hairy pretty quickly. Time to get educated!

MrFTBN, you might like to try out Spectro, a VST/AU plugin for audio software that effectively lets you photoshop the spectrogram in real time while the sound plays. It's fun, and free to try.
posted by moonmilk at 7:43 PM on April 2, 2011


I wonder if this or similar technology is the basis for Melodyne DNA ("direct note access"), sw that isolates individual notes in a mixed track. I expect this guy is going to hear from some of the music plugin companies if he hasn't already. It's very cool stuff. How long will it be before we see it put to practical use: the turd-polishing that lots of other audio plugins are used for?!
posted by keys at 10:40 PM on April 2, 2011


the turd-polishing that lots of other audio plugins are used for

Here's my favorite.
posted by flapjax at midnite at 10:50 PM on April 2, 2011 [2 favorites]


What I recall from Melodyne's press release is that it picks out the frequency content that belongs to a particular sound source by correlating the tiny little wobbles of volume, pitch, and timing that are shared across a source -- if my fundamental has a certain vibrato, then so do all my harmonics.

To cleanly pull out a particular instrument or voice, presumably this has to do something similar. I mean, it's pulling out the whole guitar sound, not just the parts of it that happen to resemble someone grunting. The key difference being that the user supplies something to correlate it with, instead of just correlating it against itself.

Also, I am thrilled by the dynamic range on "Every Breath You Take." Look at that. Gorgeous. Suck it, compressors.
posted by eritain at 11:06 PM on April 2, 2011 [1 favorite]


Also also, that de-noising stuff dearly needs to be put into cell phones that train on their primary user's voice. Holy toledo that would be useful.
posted by eritain at 11:13 PM on April 2, 2011


I expect this guy is going to hear from some of the music plugin companies if he hasn't already.

The video starts with "Adobe Inc" and the graphics look a lot like Soundbooth.
posted by odinsdream at 6:40 AM on April 3, 2011


The video starts with "Adobe Inc" and the graphics look a lot like Soundbooth

Actually it looks like the "adaptive noise reduction" feature in Adobe Audition (feature-wise; I haven't compared the graphics. But Audition is the more "pro" of Adobe's audio editors).
posted by keys at 8:13 AM on April 3, 2011


« Older At the height of the 2008 banking crisis, Antonio ...  |  Revolutionology is the blog of... Newer »


This thread has been archived and is closed to new comments