[ELECTRON] Idle speculation about music and compression files

Gordon JC Pearce gordonjcp at gjcp.net
Fri Apr 2 16:43:57 UTC 2010


On Fri, 2010-04-02 at 12:42 +0100, chris wrote:

> Does anyone have ideas about whether it would be possible to make a piece
> of music that would be impossible to store digitally, compress as MP3, etc.?
> 
> I have this foggy notion that the compression voodoo loses certain types of
> sounds, at the top and bottom limits, and it's therefore feasible for a
> tune (loosely defined ;p ) to completely disappear if that's where all of
> its action is.

Most lossy compression uses perceptual effects to determine what can be
thrown away - ie. if a particular frequency cannot be heard because it
is masked by another, louder frequency nearby it will be discarded.

Now, as you increase the amount of compression, you become more ruthless
about what you throw away - so the amount of detail remaining decreases
and the reproduction is less accurate.  To see this at work, take a
recording of some nice bright TR-808 hihats, and make several compressed
copies of it at lower and lower bitrates.  You'll hear that at 320kbps
it sounds almost unchanged, at 160kbps you have to compare it with the
original to tell the difference, at 128kbps it will sound a bit
"sizzly", at 96kbps there will be distinct "slurring" on the attack
portion and at 16kbps it will sound like someone shovelling sand into a
bucket.

Okay, so can we make a piece of programme material that would be
destroyed by lossy compression?  Well, you could make something that was
deliberately designed to put the information in the bands that would be
dropped out - but the whole point of dropping bands out is that you
wouldn't hear them anyway!  So the chances are that you'd only be able
to do this if it was a piece of music intended for watching on an
oscilloscope.

It's interesting to note that although simple AFSK data (like synth
patch dumps recorded to tape) can be Ogg- or MP3-encoded and played back
successfully, there are data modes that rely on phase-shift keying and
the relationship between phases in a signal that are extremely badly
affected by lossy compression.

Gordon MM0YEQ




More information about the members mailing list