Originally posted by redAPPLE
isn't it a fact, if one converts .cda (normal cd files) to .mp3, it removes data that is not audible to a normal person.
i mean with audible, data that the human ear rarely hears.
that is the reason why .mp3 files are a lot smaller (well compression helps too).
but if .mp3 files have already been "compressed" and data removed which is not necessary, which part would still be removed to make .aac files a lot smaller.
maybe it is the compression, right?
MP3 works by approximating the sound waves, not by just "dropping" irrelevant sound.
A really bad analogy: Imagine lookng at a a wavy line function on an x/y axis. Now, you could go through and say, "at x=1, y=37; at x=2, y=35", etc. That's sampling. That's like PCM/WAV format audio.
You could also look at the line and say, "between x=0 and x=500, it resembles the equation y=38-x^2, and then between x=300 and x=800 it is more like y=x^3-17*x^2," etc (overlap intentional). That's more like (kinda sorta
) what MP3 does. It approximates to get a wave form that is "sorta" like what you started out with. And, yes, the standard defines what kinds of approximations are allowed to be made by how noticable such an approximation will be to the "average" user's ear. In standard MP3, the bitrate is defined, so you won't get the same "quality" of approximation amongst all bits of a song (for instance, pure silence will be very well approximated, but will take up just as much space per second as the flight of the bumblebee which is noticably lacking in accuracy); in "variable rate" MP3, the quality standard across any frame is kept consistent, and so the bitrate (bytes/second) required to meet that quality standard will shift from frame to frame.
AAC employs the same principles as MP3, but allows for more involved approximations. For instance, perhaps the two above approximations could be expressed more precisely with a single x^4 order function, and perhaps an x^5th order function would give even more fidelity compared to the original. Also, precisely what is "audible" to the human ear has changed between the standards, such that much of what MP3 dropped AAC keeps preferentially because it has been found that, yes, the human ear
can distinguish the difference.
Note that the above functions are completely bogus; real MP3 approximations would involve sines and cosines, not x^n functions, as sound samples tend more towards sine/cosine waveforms than arbitrary x^n functions.
