Stayin' Alive in MIDI hell


You must mean this right:


On the other hand, it seems slightly easier to play than Rachmininoff.


You watched Pink Flamingoes on that thing?? Gad.


It was done on a physical piano here:

I think the idea is that you take the original audio, run it through an FFT to get the frequency spectrum at each timeslice, then quantize that spectrum to the 88 piano notes + relative amplitudes.


Ninja’d me. :smile:

Yeah, it is a kind of quantisation of the speech formants. If I’m reading Ablinger’s description correctly, it looks like he used a bank of band pass filters centered on semitone intervals to derive his MIDI data: - (Check “More technical details”.)


Yes, I would imagine that IS what Hell sounds like. Then again…I would say that about most MIDI tunes.


Okay, maybe I’ve been out of touch, but how has that not ended up being used in a haunted house movie yet?

Damn, if I were a particularly ambitious Phantom of any well-appointed opera house, and haunting the piano was my only way of communicating, you can be bloody well sure I’d do it.


Not wavelet. FFT. It’s basically making an 3D graph, x axis is time, y axis is frequency, z axis/brightness is intensity. Then basically painting the piano roll on top of the bright spots. Bam, you’ve got a piano roll that roughly matches the whole signal. If you have enough dynamic range, and can bang the piano fast enough, you can make it sound like pretty much anything.


It is usually FFT, but you can use wavelets as well. Or even Walsh functions, if you’re feeling a bit mad!


I’m sure you are right in the sense of that’s-how-they-implemented-it. I was thinking that, if you wanted a really good fit using the piano, you would need to use a transform that acknowledges that the piano doesn’t play pure tones. (Like FFT implicitly does; but computing the FFT is much easier because of it.)


Putting Conlon Nancarrow’s studies for player piano through an audio-to-MIDI converter would certainly not improve upon them. But performing a reading about him while vocoded through this would be a blast.


More than one way to skin that cat. Ablinger’s was evidently static analysis done with a series of digital bandpass filters at semitone intervals.

closed #33

This topic was automatically closed after 5 days. New replies are no longer allowed.