WAV FTW OMG: Why mp3s suck for DJs
Once again, pursuant to all the jibber-jabber on good old rave.vic, I am here to spread truth on the realities of digital. I would warn the non-technical that there is SCIENCE involved in this discussion, so be prepared, all ye sinners! Bring torches and rioting crowds! For today I am going to tell you why MP3s are bad, bad, bad.
As you may or may not know, MP3s sound mostly as good as full-bandwidth WAV files, and take up substantially less space. I just encoded a six-minute track that was 68 MB as a WAV, and as an MP3 it was 14 MB. So the compression ratio is about 4:1. But where on earth does all that extra data go?
I’m *so* glad you asked.
MP3 compression is based on a simple theory: What They Can’t Hear Won’t Hurt Them. It uses a psycho-acoustical and physical model to determine what frequencies will not be heard by the innocent listener, and then downsamples the heck out of those frequencies. And it does a generally great job of it. I currently have Winamp on shuffle, going between WAV and MP3 version of a remix I’m working on, and I don’t think I could tell the difference. (though I just tried, and was two for two. Huzzah!)
HOWEVER. When you look at the exact methods that mp3 compression codecs use, some scary shit comes to light. Here come the (simplified) science: The compression algorithm takes 26 millisecond chunks of an audio file, and splits them into 576 frequency bands, from 20 Hz (which is bass so damn low you can’t hear it) to 22050 Hz (which is the shrillest of high noise. You probably can’t hear it either).
Our compressor then decides, based on which of those frequencies is most likely to be heard by the lucky listener, how high-quality each band needs to be. It downsamples as needed, and then moves on to the next chunk, finally doing some cleanup for aliasing at the end, and then gives you a bright and shiny .mp3.
There are two problems with the above method of encoding. One, human hearing works differently at different volumes. To be specific, very low and very high frequencies are harder to hear at lower volumes. This is why some records sound great in a club but sound like pee on your sissy home speakers.
So, if your mp3 codec assumes playback at 80 dB, and downsamples things accordingly, when you play the song back at 110 in a club, low-fidelity parts will jump out at people. I would stress here that I don’t *know* what playback level the LAME codec assumes – if anyone knows, please tell me. With that said, it’s still an issue. Probably not a huge one though.
The much-huger issue is the way in which compression codecs divide up those 576 frequency bands. They do it in a linear, rather than exponential fashion. For those of you who did not have a lightbulb go on, please allow me to explain. Again, look out for SCIENCE.
Go to your piano, or keyboard, and play the lowest darn C that you can. Then play the C an octave up. What’s the frequency of the second C, relative to the first, lower C? It’s 2:1. Now go up another octave. What’s the ratio to our first C? 4:1. How particular.
The upshot of this is each octave you go up doubles the frequency of a note. So let’s pretend that our first C is at 60 Hz (please note that 60 Hz isn’t actually a C). The next octave is 120 Hz. Then 240 Hz. Then 480 Hz, then 960 Hz. That’s FOUR OCTAVES of music. But our codec, because it assigns frequency bands linearly, only gives those all-imporant low frequencies about 24 bands. The other four and a half octaves? They get around 552 bands. Well then.
This is a great scheme in some ways – it focuses detail on the areas in which human hearing is the most exacting. (1500 to 6000 Hz or so) However, it shortchanges the low end that is an inextricable part of house & techno. This, kids, is why you have to turn up the bass on your MP3s. Because, honey, it just ain’t there. Be a man, spend the extra $1 at Beatport or Bleep, and buy a WAV or a FLAC. You’ll be glad you did.
(Details were taken from this article. If you have another source that disputes or extends this, please let me know – the topic is one that interests me very much.)