The Podcast Piano
October 9, 2009 10:22 AM
Inspired by Peter Ablinger's speaking piano, and featuring three familiar voices.
This is an attempt to recreate Ablinger's uncanny "phonorealism" effect, which can make a piano sound like it's speaking words. After reading the thread it occurred to me that I had a pretty good idea of how it worked, and wanted to see if I could pull it off.
Here are the (mixed) results! I used my favorite excerpt from a recent Metafilter podcast, for bonus self-referentiality.
I ran the excerpt through Max/MSP and the sigmund~ object to analyze the audio, which is then "pixelated" and sent to a General MIDI piano sample. I preserved the original panning so you can (sort of) tell who's speaking depending on where they're coming from.
Now, who wants to build me a player piano?
This is an attempt to recreate Ablinger's uncanny "phonorealism" effect, which can make a piano sound like it's speaking words. After reading the thread it occurred to me that I had a pretty good idea of how it worked, and wanted to see if I could pull it off.
Here are the (mixed) results! I used my favorite excerpt from a recent Metafilter podcast, for bonus self-referentiality.
I ran the excerpt through Max/MSP and the sigmund~ object to analyze the audio, which is then "pixelated" and sent to a General MIDI piano sample. I preserved the original panning so you can (sort of) tell who's speaking depending on where they're coming from.
Now, who wants to build me a player piano?
posted by speicus (7 comments total) 3 users marked this as a favorite
Nice. I could definitely hear the "Oh, really? Well, yeah..." and the "Oh, cool."
posted by umbú at 11:27 AM on October 9, 2009
posted by umbú at 11:27 AM on October 9, 2009
suggested improvement: have three sigmund~ objects, with a lowpass before the leftmost, a series of bandpasses, and finally a highpass, and send each of their outputs in parallel, because many vocal sounds contain parallel simultaneous frequency peaks.
posted by idiopath at 2:18 PM on October 9, 2009 [1 favorite]
posted by idiopath at 2:18 PM on October 9, 2009 [1 favorite]
err. by three I clearly mean N signumd~ objects, and N-2 bandpass, a lowpass, and a highpass. With the right bandwidths / spacings I think you could get much clearer speech.
And what you have already is nice, and in places quite intelligible.
posted by idiopath at 2:20 PM on October 9, 2009
And what you have already is nice, and in places quite intelligible.
posted by idiopath at 2:20 PM on October 9, 2009
idiopath -- Thanks for the suggestion. Correct me if I'm wrong, but doesn't sigmund already output parallel simultaneous frequency peaks? I'm not using it as a single pitch tracker -- I have it tracking 32 peaks in order of magnitude per channel (probably more than necessary, as the last few peaks hardly ever sound). It seems like your suggestion would have the same result, or very similar, unless I'm not following.
I think the main things making this less than intelligible are the "resolution" of the grid (i.e. time between attacks) and the fact that it's 3 voices with a lot of crosstalk. I managed to get a much more intelligible result with a single voice recording, which I'll post tomorrow. Stay tuned!
posted by speicus at 4:34 PM on October 9, 2009
I think the main things making this less than intelligible are the "resolution" of the grid (i.e. time between attacks) and the fact that it's 3 voices with a lot of crosstalk. I managed to get a much more intelligible result with a single voice recording, which I'll post tomorrow. Stay tuned!
posted by speicus at 4:34 PM on October 9, 2009
Yeah, never mind, you are right, I had only used sigmund~ in single pitch tracking mode before, but I finally checked the help file and indeed it can do parallel frequency peaks, my bad.
posted by idiopath at 5:07 PM on October 9, 2009
posted by idiopath at 5:07 PM on October 9, 2009
Hey! I've got a player piano! Get in touch - we should try to make it go!
posted by moonmilk at 5:23 PM on November 5, 2009
posted by moonmilk at 5:23 PM on November 5, 2009
You are not logged in, either login or create an account to post comments
C: Remember when we were talking about that dream I had about sex with unicorns?
J: This one time...
C: This ties back to that.
M: The rainbow stuff will clear out.
C: So anyway, the penicillin should take care of it.
J: Ha!
M: That wasn't the unicorn, that was the bus terminal.
J: You know, everyone has Herpes. That's what I learned from Ask Metafilter.
C: Yeah, pretty much everyone does, I guess.
M: Oh, really? Well, yeah...
C: But, anyway, there's this recording by, I have no idea how to say his username, it's another adventure in pronunciation, but...
J: Spykus!
C: Is it Spykus?
J: Spaykus? Specious? Spaykoos? Ask klanglangston. They're friends.
C: Ok. I'll ask him when I'm in LA.
J: Isaac, Isaac is his real name. He's nice.
C: Yeah, actually it's funny I know his name is Isaac but I've never tried to deal with his username. He was on the Metafilter compilation album, and this song actually, the original version of it, was on the compilation. And now, like two-and-a-half years later, he posted, and this was back in May but I don't think I mentioned it at the time, he posted an updated version of it and I really liked it when it was on the album and I really liked what he did with it for the redo. So I just wanted to throw that out there.
M: Like a complete rearrangement, or just a better version?
C: He just sort of laid on some more layers and cleaned some stuff up. Filled it out so it's a little bit more full of a sound than it had at the time on the album.
M: Oh, cool. I guess we will end the podcast with this and play it out. Play us out, "Distance of the Moon Remix"! By Speeeekus...
C: Cat.
M: (Keyboard cat noises.) That wasn't me, that was my cat on my keyboard doing it.
C: Tell the cat to be quiet, we're recording a podcast.
M: That's right. (Keyboard cat noises.) My cats are always doing that to me, playing me out of rooms.
posted by speicus at 10:25 AM on October 9, 2009