Music File Compressed 1,000 Times Smaller Than Mp3

The virtual clarinet is obviously a large file

It's a large file because it's a wav recording of his computer simulating the clarinet. You only need to get the player once, just like media player, and then the musical scores are tiny files.

But I agree, it has a long way to go, I could quite easily make out the virtual one. Interesting work though.

That's exactly what I thought.

Sounds like they're reinventing the wheel.

MIDI already can represent note value, velocity, aftertouch, modulation, etc. There have been MIDI breath controllers for at least 25 years.

A lot of the major multi-track audio packages (Logic, ProTools, Cubase) have had an audio-to-MIDI conversion feature for 5 years or more.

Yes, this is basically just an advanced form of MIDI.

I don't consider this "compression" of a music file, it doesn't even take a music file as an input. It is limited by what instruments it can accurately reproduce. It will never be able to reproduce vocals, for example.

It's kind of like saying sheet music is a compressed form of a music file... it's not, it's the music described in a completely different format altogether.

This is amazing, and I think it is a form of compression.

It is no different than a translated file (mp3) triggering a player and reproducing a piece.

It is no different than a translated file (mp3) triggering a player and reproducing a piece.

Actually, it's much, much different than an mp3. MP3's work by using FFT (Fast Fourier Transforms) to look at small slices of audio in the frequency domain. Then the frequency components which are not as prominent as others are discarded, resulting in a smaller file size (and, often, with noisy source material, audible artifacts).

This works more like MIDI (as previously noted), where pitch and other characteristics of the source audio are analyzed and stored as data.

I suppose you could say that both techniques are forms of data compression, but they take completely different approaches. With mp3, *any* source can be compressed via brute force, while the technique under discussion relies on more sophisticated analysis.

You mean like a song by T-pain?

I also suspect they picked this instrument on purpose.
it would be interesting to see how well they do with something with a bit deeper pitch.
I would find it hard to believe they could represent the body of a base note with only 50 bytes a second.

we should also beware.
They once told us cd's sounded just as good as vinyl and lasts forever. Neither turned out to be true.
They told us that mp3's were perfect representations of cd audio.. except they arent either and the more you compress them, the less like cd they are.
we keep making these consesssions to "almost perfect" technologies and one day we will be back at am quality or 8bit sound... ok maybe not but audio technology appears to be reversing direction and yet they still call it progress,

ok maybe not but audio technology appears to be reversing direction and yet they still call it progress,

I get cynical as well, Joules, but there's also cause for celebration.

On the home-recording front, high-quality (24-bit, 192KHz sampling rate, uncompressed--much better than CD) audio interfaces are getting affordable (8 channels for $400).

Apple offers their lossless codec for ripping CDs to the iPod. It truly is (mathematically) lossless. While I believe Apple does not sell uncompressed music, on archive.org, one can download SHN or FLAC-encoded (two other lossless formats) audio of many fine musical acts.

On the iTunes Music Store (iTMS), one can buy "iTunes Plus" music encoded at 256kbps AAC--much better than 128kbps mp3.

My understanding of this is that they are basically combining something similar to MIDI technology with synthesized models of the various instruments. When these two are combined, they have both a very small file containing the note data, as well as a fairly small file containing the "instrument" data. Since this whole way of doing it is pretty new technology-wise, it is impressive.

It's not even really a new concept though... parametric (as opposed to sampled) modeling of instruments, known as digital waveguide synthesis, has been around for decades, and available in consumer level MIDI implementations as early as the EMU8000 processor on Creative's AWE32 in 1994. (I was hugely interested in quality MIDI reproduction back then... this was a few years before MP3's hit the scene. Anyone remember the Gravis Ultrasound, or Timidity?)

Not to knock what these guys have done, they've clearly advanced the sophistication a great deal and I'm very impressed with where they're taking it... but I think this is more "evolutionary" than "revolutionary".

If it works as advertised. i have a feeling it might only work on instruments that are strummed, and not with ones which have all the notes played simultaneously.

It will still be cool to get Jimi Hendrix to play on my tacks though.

the video on their website is pretty amazing, but it could be smoke and mirrors, and a lot messier in real life. even so, I make noisy experimental music so that could work to my advantage.