Why MP3 converted files are unlikely to vocal eliminate
In case you are interested...
When a PCM (Pulse Code Modulated) linear audio file that was mastered in true stereo is converted from PCM to an MP3 compressed file, to my understanding, there are changes made in both channels that are not identical. Each channel is processed with the compression algorithm separately.
Like all psychoacoustic compression algorithms, MP3 removes audio frequencies and other components that are not "perceived" by the human ear. However, this changes the resulting bits in a 16-bit audio sample (or digital word) from what the original file contained. If it didn't change, there was no information removed. It is clear that MP3 definitely removes material to bring about the data reduction (128Kbit MP3 is 11:1 compressed).
Now, in the vocal elimination process, it is very critical that the true stereo channel data be exactly preserved. MP3 compression changes this data, and thus reduces, if not altogether prevents, the effect of vocal elimination. The fact that Cid_cat has found one MP3 file that had
"exceptional results" is very surprising. In this case, the processing of both channels must have been near identical.
For vocal elimination, the words in both channels for the frequencies of the voice (and any other sounds/music that is center panned) must be inverted, channel swapped and summed. When inverted, they become the digital opposite (or negative in mathematical terms). When two identical, digitally opposite words are summed, the result is zero. Thus, if one channel's sample word is changed slightly in the MP3 compression process, when inverted and mixed with the other channel, there will not be an exact cancelling to zero. What is left is a highly variable artifact, that is in reality, the difference between the MP3 processed audio samples from the two channels.
|