The first few seconds of the Rush song "Subdivisions" consists of a low synthesizer chord, repeated a few times. For some reason, if I use any of Dibrom's tweaked --alt-presets (standard, extreme, or insane), the first attack sounds "rumbly".

If I use the ABR --alt-presets, r3mix, -V 4, or -b 128, there are no problems. Also, it's fine in OGG and MPC, any quality level.

I know it can't be a VBR issue, since the problem appears in "insane" mode as well. Anway, here's the file.

Aw, crap - there is no problem with the file. I use the WinAmp equalizer to boost bass frequencies (I know it's not the greatest, but I've been pretty happy with it). As soon as I turned off the eq, the mp3s encoded with --alt-preset standard, extreme, or insane sounded fine.

I guess a follow-up question is why do these presets have this effect on low frequencies, where other presets don't?

In short, the problem lies in WinAmp's crappy EQ. I just made the same mistake a few days ago. Please see this thread for further detail.

Maybe alt presets preserve bass, so that they are clipped when you boost them.
That's a good example of why Naoki Shibata didn't allow boosting any frequency in his super equalizer plugin.

The alt presets use nspsytune, which is known to have a rumbly bass at lower bitrates (around 128 to 160 kbit/s), which is typically just covered up in ABR/CBR by using --ns-bass -xx. In addition, mp3 in general sometimes has a problem in the first couple of seconds. Add your equalization, and you've just handed it a triple whammy.

ff123

Thanks for the info - now, are you implying that using ns-bass with Dibrom's VBR preset would be a bad idea?

Also, r3mix uses nspsytune and doesn't have the rumble, so I wonder if there's something else going on here...

Pio2001, I don't get this rumbling from the original wave file, which is why I think the VBR alt-presets are increasing (or otherwise altering) bass

When you are using any sort of post processing (including an EQ, especially when boosting frequencies) on a file psychoacoustically encoded, you run the risk of breaking the masking used. This is exactly what is happening here. There's nothing wrong with the --alt-presets if they work the way they are supposed to and sound the way they are supposed to before the post processing.

As for other presets which may or may not be subject to this after post-processing (I'd have a feeling --alt-preset fast standard might be the same as --r3mix), it probably means they are not using masking as efficiently (introducing quantization noise in as tight of a threshold). This should be considered more of a bad thing than a good thing because it would usually result in files with less of a size/quality ratio.

Could it be just clipping ?

Actually that could be possible (though I'd be surprised) I guess since --r3mix uses --scale and the --alt-preset vbr modes don't.

One would have to encode with --alt-preset standard vs --r3mix --scale 1 and --alt-preset standard --scale .98 vs --r3mix to test it.

Thanks for the feedback - this has been really enlightening.

This brings up an interesting point - many people like to listen to their encoded music with equalizers; now even if they use a good equalizer that doesn't itself degrade the quality, it will break the various masking assumptions made in the psychoacoustic model. Is there a way to account for this in encoding (i.e. a "less suspectible to introducing problems when equalized" mode that tries not to drop any one frequency range too heavily even if it normally would), or is the solution to just use higher quality settings so that "exposing" a certain frequency range like that still won't sound bad?

Well, first off I'm not 100% sure that is what is causing this (could be clipping as Pio2001 said, I haven't checked yet), though it's probably likely.

Secondly.. I don't think it's practical to try and account for this in the psymodel. It's really not the encoder's job to try and guess where people might want to boost or cut certain frequencies after the fact. If you try and do something like this, you're going to end up with a much less efficient and clean psymodel (if you could even do it effectively at all -- how do you know what kind of eq curve the listener is going to use?). To answer your second question, you could get around this to some extent by using less masking, but that means larger files (and some people already think --aps is too large).

I think this is really one of the catches of this type of lossy compression... it's meant to reproduce the original signal in an unmodified environment. If you need to do a lot of post-processing, you should be using lossless.

That being said, I think this problem might not be so evident if you don't boost particular frequencies, but instead lower others and just increase volume to get a similar effect. This is kind of like what Pio2001 was saying about the supereq.

Also, I'd suggest any further testing/discussion on this topic in relation to eq's should be done using Naoki's supereq and not the winamp one, since it's very far from ideal, and could be a large part of the issue itself.

Dibrom - thanks for the reply. To clarify, i wasn't talking about this particular clip or the Winamp EQ, but just the general problem of "feeding lossy-compressed music to an EQ." I don't really have much to add in reply to your points; seems to make sense to me. While lossless is of course the ideal solution if you're going to be doing a lot of post-processing, it seems that some lossy compression might still be acceptable - i.e. if all you're doing is EQing (and not pitchshifting or something) you can still safely lowpass at 20 KHz (and probably lower), and you can probably still do quite a bit of masking.