Music File Compressed 1,000 Times Smaller Than Mp3 | Hacker News Reader

Music File Compressed 1,000 Times Smaller Than Mp3

sciencedaily.com

10 pointsnreece16 comments

Sounds like this is a little bit more heavyweight than MIDI. MIDI uses simple events and relatively simple representations of sound to play back compositions. A MIDI file contains no sounds- it is more like a sheet of music. You can play this file using an instrument you choose.

The compression they speak of is really just a more efficient storage of data. Memory that was once used to encode raw sound data is now used to encode information about sound dynamics- totally different from MIDI.

I would like to know how large the libraries are though. Seems like they could easily take up a lot of space. Still a very useful technology though if they can perfect it. Can you imagine what streaming music would be like?

Well it is a little bit closer to old school .MOD files (anyone remember those?) than MIDI. This kind of compression can be useful for a very narrow field of music but not for general type.

I thought the sample sounded OKAY, not great. The article certainly made it seem like this has a lot of potential though.

The biggest thing missing was the ambient white noise and transitions between notes- two things that are more commonly found in music played from a physical instrument. I think that is their biggest remaining challenge.

If you have purely synthesized music then you might be able to make the difference between the original and the compressed imperceptible.

Voice synthesis technology exists as well that, if perfected, would make it possible to generate music and lyrics entirely from the command line.

Then again in the 1980's some people predicted we would have flying cars and colonies on the moon by now so it could just be hype.

1) A place called "Science Daily" using the phrase "x Times Smaller Than"... arrrgh. (offtopic: Google "times smaller than" in the next couple hours to get a peak into how Google seems to be ranking sites... I was surprised to see this topic was 60% of the first page)

2) Outbound links! Please, people, this is what the internet is about!

3) This is an interesting article. Unfortunately because of a lack of (2) it's hard to tell if they're making an effort to do this in a way that supports a wide array of instruments (an instrument synthesis API would be pretty cool), or if this research has a more limited scope.

I admit I don't know enough about sound and music to speak intelligently about what this research could do, but it seems like they could abstract it to multiple classes of instrument and open a whole new field of synthesizing instruments that sound like they should exist but don't. I'm thinking of the sound equivalent of the Memristor -- the (1940s?) math and physics showed that it is possible but nobody had made one until >2006.

An elegant demonstration of the probability-theoretic truism that to compress information is to understand it and vice versa. (Of course this doesn't work with shallow compression like zip.)

so to recreate, say, a beatles song, you'd need to model all the equipment at abbey road, all tge guitars and drums and amps and mics and exact positions and cables and consoles and tape etc etc... and once you have these terrabytes of data, the song is only a few kb. uh...

> ... and once you have these terrabytes of data, the song is only a few kb. uh...

A few kb seems like overkill... I reckon you could get it down to 1 bit.

So, I hear there are these things called libraries... but not the book kind! Are they some sort of new scienc-ey development I haven't heard of yet?

If you manage to recreate the physics of the members of The Beatles, a bit should suffice. It might even get you the hordes of screaming teens.

Ok, so some researchers made a 20 second clarinet piece fit in less than 1kb. How is this not an overly complex reinvention of MIDI?

Because as far as I know, MIDI uses pre-recorded sounds to synthesize tracks. This, on the other hand, simulates the physics of the given instrument, which lets the program create much more complex sounds. This has the type of potential that MIDI cannot have in anything resembling its current form.

Let's set aside for the moment the issue of whether it makes sense to call this "compression". What I'm interested in is: does the simulated instrument sound any good?

At first, I was extremely impressed... but then I figured out I was listening to the original. Listening to the simulated version, it really didn't sound any better than, say, the "clarinet" mode on a cheap Casio keyboard.

PS. I used to play the clarinet. I don't know if this makes me pickier about clarinet sounds than other folks.

It's not just you. The compressed version was mechanical and even out of tune, although the crescendos sounded about right.

It's obviously a waste of time as a recording format, but it could have applications in decoding instrumental performances -- maybe helping students to see what they are doing, or providing some quantitative insight into great performances.

Isn't this an April fool's joke?

Misleading. The file is not actually a recording, but a set of parameters to drive a simulation. It's really just a glorified MIDI file.

So this is like TechCrunch for umm.. science and stuff.

In other news (on the same site): "Listening to personal music players at a high volume over a sustained period can lead to permanent hearing damage, according to an opinion of the European Union Scientific Committee".

Wow, who would have thought!

Reply on news.ycombinator.com