[Nordenstam]

Originally Posted by greggybud

As MP3's become more popular does it warrant it's own mastering?

Depends. The typical ways of gaining loudness in CD mastering does little good for psychoacoustic coders. So in that case, I would say yes. If the CD is done without severe limiting and is technically correct level-wise in every possible way, it'll translate fine to lossy coding.


Been testing and reading up on this a bit and the impression so far is that lossy coders doesn't like modern CD mastering. Reasons seems to be both related to the way the coders work, and the way digital signal levels are treated and measured. What I've found so far is basically:

Coders divides the spectrum into frequency bands, resembling the critical bands of our own hearing perception. Within each band and somewhat to neighboring bands, loud frequencies will mask other frequencies. The basis of audio data reduction is to remove these signals we hopefully won't hear anyway. Different parts of the spectrum is given priority depending on what the psychoacoustic models believe humans hear. The rate of success depends on the coding models and the signal being coded.

This bit of theory suggests that limiters and jam packed audio may not be the best starting point for the coding. They may be tuning them for the typical audio of today, though I have a hunch this is not the case. It seems they are set for fidelity in vocal recordings and normal instrument sounds. Audio signals have harmonics that define the character of that signal. Distortion, often from cliping and limiting the signal in mastering, will add spurious information. This may gain priority in the coding process, effectively lowering the fidelity of the more important original signal. In any case, less limiting gives 'more space' around each sound, making the guesswork of the coder easier. To me, 'breathy' music sounds better when coded than massive masters. The text book facts seems to support it, but I dunno what really goes on in the coder.


Less subjective and perhaps more important is the sample point peak levels and the intersample peak levels(link to explanatory paper).

The sample points themselves will change in the coder process. The output of the decoder is a synthesis. Although it will sound much the same, the actual waveshape is not an image copy of the original. The peak points will move around a bit and probably end up with a slightly higher level. Maxing out at zero dB flat on the CD is guaranteed to result in clip'ing in the decoder process.

How much the peaks will change depends on the nature of the signal. If sample point peak level and intersample peak levels are nearly identical, the change will only be a few tenths of a dB. If the sample point peak level are far below the real intersample peak level, the effect is more severe. It also depends a lot on the selected bitrate, with lower bitrates gives more discrepancy. The coders are strickt low pass filters, any extraneous high frequency information will be removed. The result is to act somewhat as a reconstruction to 20-20K bandwidth on the waveform.

Limiters and cliping creates higher frequencies where the peaks have been squashed. Removing high frequencies smooths the flat peaks to softer waves with larger peaks. A typically extremely loud modern master with peak clipping often exhibits several dB's higher peak level from the lossy coding process.

The way to do it is to ease up on the flat squashing, rather pushing the bottom up if massive is what the clients want. And work with strickt control of the intersample peak levels.

Combined with the absolute need for lo-fi compability and zero dynamic range of headphone listening in loud outdoor ambient noise, it's a real bugger to master!

1. As a standard procedure, do you as a mastering engineer convert each song to MP3 and listnen?

Only when mastering for Mp3. Would love to have a plugin that does coding and decoding in realtime! For checking the end result when crunched through a codec. AFAIK, there is no such thing that caters for this specific need of mastering for mp3. Going offline to convert back and forth between Mp3 and wav formats is a long way around.

2. Over the past year it seems I'm having an increasing difficulty detecting differences in MP3 from .wav when played in my studio. ... And what should I be listening for to detect the differences?

Try converting to mp3 and back to wav, line files up to do seamless A/B switching between the wave file and the mp3. Also, line the samples up so the match accurately and try phase reversing one of the channels so you only listen to the difference product between the original and chrunched file. Should give you a clear indication of what to listen for in the Mp3. Creaky highend, broken glass" artifacts, flanging/phasing sensations.

Originally Posted by Timecode

I recently did a test on the new Rattatat album between a 320 Kbps rip and the wav - I could hear the difference fairly cleary, but it seemed to me that the mp3 was slightly louder, like maybe 0.5db RMS or less. I know the encoding can give you clips by changing the transients, but does it tend to boost RMS as well?

Using the lame encoder, my experience have been the opposite. RMS tends to measure a tenth or two of a dB lower than the original, while the peaks measure higher.



Regards,

Andreas Nordenstam


Disclaimer: I'm just a learner in mastering. Am thankful if someone more proficient will correct my mistakes if needed!