Most people master at -0.1db or -0.3db to avoid problems with intersample peaks during MP3 conversion.

Right now I'm using FabFilter Pro-L limiter which has an intersample peak detector that shows you the clipping that will occur when you convert to MP3. If you turn on the meter, it will show peaks of +2db or above (with the oversampling feature turned off).

Does the meter suck? If it's right, it's quite pointless to master at -0.3db when you're still going to have 1.7db of clipping in your MP3...

the ISP meter on Pro-L is good...and the 4x oversampling takes care of the ISP so if you set the ceiling for -0.3dB, that is what you will get...within < 0.1 dB usually, so there should be nothing over 0 dB...

Well but let's say you use a limiter that does not have the oversampling feature. Is it possible that you get 2db of clipping when you convert it to mp3? If that's true, there's no reason in bouncing at -0.3db...

well, you do get the ISPs in the mp3, they do not go away...
they do not get "clipped" off if that's what you mean...

You only get them in the mp3. But 2db?

People bounce at -0.3db to prevent problems with mp3 files. You're only preventing clipping if the ISP meter shows something from +0.1db to +0.3db, but if it clips by 2db (and that's what the meter shows), mastering at -0.3db is pointless, you should master at -2db to make sure there's no clipping in the mp3. But nobody masters at such a low level.

i'm not sure i follow you on the "only" part...
if the ISPs are in the .wav (source audio file), they are in the mp3, too...

Man, sorry if I seem rude but... do you know what ISPs are?

There aren't any ISPs in the wav file. ISPs are caused by mp3 conversion (mainly). The meter shows the ISPs that will be present once that wav is converted into an mp3...

ok, i did not mean that the ISP are literally "in" the .wav file, i guess that was careless of me to say...they are created at the DAC from the .wav (.aif whatever) file...

Yes that's right. If the ISP meter shows +2db, it means that when that wav is converted to mp3, that mp3 is going to have 2db of clipping.

My question is: is it possible that intersample peaks can get up to +2db? Shouldn't they stay under +0.3db or something like that, since -0.3db is pretty much the standard output level when mastering?

The point of mastering at -0.3db is to leave some room for ISPs, but if ISPs get up to +2db you're definitely not leaving enough room.

If you master at -0.3db a wav file that the ISP meter says it's going to clip by 2db when converted to mp3, you're still going to get 1.7db of clipping on the mp3 file. Which means that exporting at -0.3db is totally useless.

What's wrong?

it doesn't matter if you are playing the .wav file or the mp3 file, you will have ISP from both files upon DAC...
if your ISP meter says there is +2dBTP, then when playing the file, regardless of it's format, before or after conversion to mp3, you will have the +2dB peaks (+/-) at DAC...

Say I'm using Pro-L and I have the oversampling feature off.
I use it on my master output and I set the output ceiling (on the limiter) to -0.3dB. The ISP meter shows something like +1.5dB. I bounce a wav and I check it's peak. It's peaking at -0.3dB, not going over 0. The MP3 I make out of it does though.

what are you using to check the level of the bounced wav file...?
if your ISP meter says +1.5dB before the bounce, and you do not use oversampling on Pro-L, then the file you bounced has ISP/TP of +1.5dB...

For example load the wav in Logic and check the mixer's level while it's playing, it never goes above -0.3dB.

Logic does not have an ISP or "True Peak" meter...it is a "sample" peak meter...
you need something like this > TB_EBU to measure dBTP (decibels true peak)

Hmm I don't think I explained myself...
Whatever IPS you might have, the wav file is still going to peak to whatever output ceiling I set. While those IPS are actually "showing up" on the mp3, making it clip. Isn't that right?

There is a basic misunderstanding here. The ISP reading in Fab Filter or any other meter is for linear PCM. It will tell you when the reconstructed waveform will exceed 0dBFS. It has no way of predicting what MP3 or any other codec will do WRT overs. For that you need something like the Sonnoc Pro Codec or the Apple tool.

Wait... so, why my wav files are always peaking at the output level I set on the limiter? ISP have no effect on my wav. Just on my mp3...

If you are on Windows then download Foobar2000 and Replaygain scan your mp3, aac/mp4, ogg, etc and then you will get a peak reading. 1.000 is 0dBFS and I believe 1.500 is +3dBFS. The lower bitrate you use the higher above 1.000 it will read. 128kbps mp3 files often read about 1.200 and somtimes higher. 320kbps are usually under 1.100.

Sent from my HTC Desire HD A9191 using Gearslutz App

Sorry but I'm testing this everywhere. My wav is peaking where I set the out ceiling, ISPs aren't affecting the peaks in any way on the wav file.

ISP have effect in both linear and encoded audio. With the ISP function turned on it monitors intersample peaks and adjusts the output to keep them from exceeding your ceiling. But now that you have a limited file with, say 0.3dB of headroom, MP3 encoding and decoding may create additional overs. IPS protection in the linear PCM world is no guarantee against MP3 reconstruction clipping.

Ok but, when I load a wav file and the ISP meter says something like +1db, that's an intersample peak that already had effect on the wav file and not the peak that will occur during mp3 encoding right?

Yeah, that's not good. You should be able to drop the level of that file and prevent the ISP (which is a product of linear PCM reconstruction) before it happens. Then save the file at that new, lower level. You'll need something closer to -1 or -2 (depending on the content) before hitting the MP3 encoder.

Download the Apple mastering tool. It's free and a great resource.

Sorry but, if i take that wav and i make an mp3 out of it, then check its peak, it says "over". Which means that those additional overs that occur during mp3 conversion were higher than 0.5db (the headroom i left in the wav). Possible? (I used the TT Dynamic Range Meter, it also tells you where a track is peaking, over is what it says on all the mp3s ive got, and it means that some clipping happened).

Definitely possible. MP3 encoding is not a linear process so unexpected things can happen. Look for a Pro Codec demo. It's very interesting to watch.

Sorry, but your understanding of ISP is wrong. ISPs have no relation to mp3. They are related to D/A conversion.

If you want to prevent mp3 clipping, ISP meters are useless. The only way to prevent mp3 clipping is to use the actual mp3 encoder to check the levels. There are some plugins that encode/decode an mp3 in realtime (as Greg mentioned).


ISP almost never reach +2 dB. On real music they are typically within 0...+1 dB. But you are really asking about mp3 overshoots, not ISPs. Mp3 overshoots depend on your bitrate and can easily reach +2 dB at low bitrates.

Greg stated good point. Different mp3 codecs will have different ISPs. Also, the AAC does have the ISPs too when the ceiling is set too high (and it depends of course on the material).

Most of limiters today have ISP protection (which does NOT protect you from codec error ISPs!) I master with ceiling about -0,4. Even with that setting mainly I dont have problems with peaks exceeding the zero.

Take a iTunes tour. 70% albums - if not more - is distorted and it is caused by encoding isp errors.

So, the ISP meter on FabFilter indicates the same overs both when I use it on my Logic project as a limiter (and it's telling me which overs will happen in the wav bounce), and also when I open I load the wav i bounced into a new project, right? But it can't tell which overs there are going to be in the mp3. Correct?

Sorry. I doubt that's the reason. 99% albums - if not more - are distorted due to heavy limiting to make them louder. Distortion you get from encoding error is nothing compared to the distortion some tracks have caused by incredibly low limiter thresholds.

ISP meters have no knowledge about mp3 overshoots. You can consider the process of mp3 encoding as adding a noise to your file at the level of -20 dB. This noise is likely to change your peak levels, but only an mp3 encoder knows what this noise will be. Same thing with AAC, OGG and other lossy codecs.

Thanks I got this, but then what is the point of bouncing a wav at -0.3dB?