I must say that I am surpised by how good audio MPEG files sound. It is also amazing how much music can be packed into 12 megaBytes of memory.
So how does MPEG compression work? Do they just discard all frequencies >15KHz or so? Or is it much more sophisticated…doesit filterout low-level tones as well?
How much of music CAN be discarded, withot affecting the intelligibility? I rememnber that some experiemnts were run by the Marantz Co, years ago, to see what level of audio distortion could be detected by most people. It turned out that most people could not tell that a pieceof music had been distoted, untill the distortion level rose to 9-10% THD (this while stereo mfgs were touting their 0.001% THD levels?
MP3 compression works on a perceptual model; it de-prioritizes or discards sound information that the human ear will be less sensitive to. It divides the audio to be compressed into frequency bands, and then considers those bands as a group of audio frames. Then it decides how much each band will “mask” the adjacent bands on a frame-by-frame basis - “masking” being determined based on a model of human hearing. From there, it decides how many bits to use to encode each frame (more perceptually important data gets more bits), given the bitrate limit you specify.
As far as how much of music can be discarded, I’m not really sure - and it’s probably dependent on the piece of audio in question.
MP3s are all the Fourier these days.
Well, I also think that it bears noting that a great number of people can tell the difference between a full-CD of music and MP3 compressions. Some differences are easy to tell, especially on the lower end with quality speakers, but on any system, the MP3’s sound a lot “flatter” than the CD.
It just isn’t the same.
That depends on the bit rate of the MP3. According to fairly thorough test of expert listeners, conducted by a German magazine, 256k MP3 sounds just as good as uncompressed CD-quality audio. Even at 128k the listeners sometimes had a hard time judging which sounded better.
http://www.geocities.com/altbinariessoundsmusicclassical/mp3test.html
From a good MP3 encoder (FhG or LAME), 256kbps is “CD Quality.” Experienced listeners using high quality audio components will, in most cases, not be able to distinguish between such MP3s and the original CD. 128kbps is considered by FhG to be “CD Quality,” because the average listener using low quality components will say it sounds “just as good” as the CD. 128kbps and higher MP3s from a good encoder should not have any audible artifacts, the audio just sounds less “real” due to the reduced fidelity and increasing lowpass filtering. Below 128kbps, audio rapidly approaches unlistenable.
Some MP3 encoders, noteably Musicmatch Jukebox and the Xing products, discard all audio data above 16Khz, no matter what bitrate you use. A good encoder will perform filtering, but it will be adaptive, with fewer frequencies being thrown away as bitrate increases.