RMAA result of a 128kbps MP3 - limitations of analysis methods

[wwenze]

Actually I did these measurements before, but I didn't realize the implications back then.

Bear with me for using RMAA, I like how it do various tests automatically.

The problem is how 128kbps MP3 measures so well. I don't think I'll get any objection if I say that the difference between 128 and 320kbps shows up easily in scientific subjective testing. I'm guessing THD doesn't include the spread spectrum effect in its calculation, but whether IMD at -70dB is audible is debatable.

The frequency response at 128kbps is pretty funky, but not as bad as transducers. Frequency response at 320kbps is fairly flat.

So two possible cases -

1) The signal is too simple for compression to degrade it by much. So while 128kbps MP3 handles this signal well it is still poor with music.
or
2) The signal has degraded much but this went undetected - no significant difference between 320kbps and 128kbps bar the frequency response. Unless the frequency response counts as a detection and accounts for the huge perceived difference between 320 and 128kbps, which I feel is unlikely.

And the respective implications -

1) There will also be equipment that handle simple test tone(s) well, but suffer with music. So what kind of testing will be sufficient? 3 tones? Square waves then compare the FFT plots? DiffMaker? (How accurate and noisy is DiffMaker btw?)
2) There can be equipment that sound bad but measure well as far as the simpler measurement methods are concerned.

Thoughts?

[DF96]

You can't test a compression method by giving it a signal which needs no compression. Given a single tone all it has to do is code it, not compress it. If it couldn't even do that well then it would be useless for music.

My understanding is that most conventional audio measurements are virtually meaningless in this situation. The whole point of MP3 etc. is to produce a sound which fools the human ear/brain into thinking that it is hearing a high quality reproduction of music.

[Ron E]

Originally Posted by DF96

You can't test a compression method by giving it a signal which needs no compression. Given a single tone all it has to do is code it, not compress it. If it couldn't even do that well then it would be useless for music.

My understanding is that most conventional audio measurements are virtually meaningless in this situation. The whole point of MP3 etc. is to produce a sound which fools the human ear/brain into thinking that it is hearing a high quality reproduction of music.

Look at the multitone signals used by stereophile - they are interesting and clearly show differences between encoders. I did my own battery of tests - I use custom LAME settings that average ~225 for mp3 files and they make me happy....

[wwenze]

Originally Posted by Ron E

Look at the multitone signals used by stereophile - they are interesting and clearly show differences between encoders. I did my own battery of tests - I use custom LAME settings that average ~225 for mp3 files and they make me happy....

Thanks for the heads-up, that prompted me to use more complicated test tones with frequencies of odd ratios, and discovering that even before compression, exporting it as 24-bit integer already creates nasty artifacts. Now I appreciate 32-bit float. I have much to learn...