Intersample peaks and output ceiling

[Alexey Lukin]

Originally Posted by stinkyfingers

but i have found that Elephant is really the best for this

Elephant's handling of ISPs depends on the chosen limiting algorithm and other settings. For example, if it's set to "Clip", ISPs are not killed at all. Oversampled limiting does not automatically imply no ISPs.

[Alexey Lukin]

Originally Posted by ProducerP

One last question... those overs that I get when oversampling is on, are they ISPs? If so, why do I see them even with the ISP meter off?

To check if clipping is due to ISP or not, switch the meter from a true peak mode to a regular digital mode and watch the levels. You may need to familiarize yourself with a particular meter, as there are different standards of clipping registration.

[ProducerP]

Originally Posted by Alexey Lukin

To check if clipping is due to ISP or not, switch the meter from a true peak mode to a regular digital mode and watch the levels. You may need to familiarize yourself with a particular meter, as there are different standards of clipping registration.

Well the manual says they're ISPs. But you said that ISPs are related to D/A conversion, what about oversampling?

[Lagerfeldt]

Oversampling will also manifest ISPs.

[ProducerP]

Well, since the limiter is clipping internally, I need to lower the out to avoid clipping, and adding another limiter after it won't do anything, right?

[Alexey Lukin]

Meters emulate D/A conversion by oversampling and predict ISPs from the digitally upsampled signal.

[stinkyfingers]

Originally Posted by ProducerP

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

you know, at first i though you were asking if i knew what they were because you thought i didn't know, but after all of your posts, it is clear now that you were asking me what ISPs are because you have no clue...sorry i didn't answer you...google might...

edit: sorry if that seems rude...

[ProducerP]

Originally Posted by stinkyfingers

you know, at first i though you were asking if i knew what they were because you thought i didn't know, but after all of your posts, it is clear now that you were asking me what ISPs are because you have no clue...sorry i didn't answer you...google might...

Don't worry I already got my answers.

[Greg Reierson]

This stuff is not intuitive. No harm in asking.

[ProducerP]

Originally Posted by Alexey Lukin

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

They do have relation to mp3.
This is what the FabFilter manual says:

The aliasing will cause inter-sample peaks and these can cause distortion later on, for example during D/A conversion or conversion to mp3.

[Alexey Lukin]

That's a mistake in the manual. ISPs are not related to either mp3 overshoots or aliasing.
It is true that a mix with ISPs is also more likely to clip after mp3, but not every correlation between two phenomena means causality. In other words, removing ISPs from your mix will not make your mp3 clip significantly less.

[ProducerP]

Originally Posted by Alexey Lukin

That's a mistake in the manual. ISPs are not related to either mp3 overshoots or aliasing.
It is true that a mix with ISPs is also more likely to clip after mp3, but not every correlation between two phenomena means causality.

Oh, ok good to know.

By the way, is the clipping that happens on the mp3 related to the ISP value on the original wav? Example, a wav with higher ISP value will result in a more distorted mp3 file?

[Alexey Lukin]

There's often a correlation between the two, i.e. if your mix has ISPs, then its mp3 version is likely to clip stronger (but see counterexamples below). The actual level of mp3 clipping is determined primarily by your bitrate setting, as well as your signal and your mp3 encoder.

Counterexamples:
1. Even if there's no ISPs in your mix (e.g. if you have used the ISP-killing limiter), its mp3 version may easily clip.
2. The mix may be full of ISPs (e.g. a 11025 Hz sine wave at -1 dBFS), but its mp3 will have no clipping (on a digital peak meter).

[ProducerP]

I just don't know if it's ok to leave oversampling off, because with the hard limiting that I'm doing, ISPs easily go above +1.5dB... Not a problem for the wav, but I don't know for the mp3. To me it sounds good... but do you think it's too much?

[Alexey Lukin]

The artifacts of mp3 are usually much worse than artifacts because of ISPs. When mastering for mp3, I would concentrate on mp3 clipping and keep it in control, not ISPs.

Don't hunt for every instance of clipping: not every clipping is audible. Make sure that you are not clipping badly.

[ProducerP]

To be honest I don't care that much about clipping... I crushed my music a lot, there is already some distortion caused by limiting, and however it's the wav that I really care about.
I was just trying to understand if such high levels of ISP on my wav were normal.

[stinkyfingers]

Originally Posted by Alexey Lukin

...2. The mix may be full of ISPs (e.g. a 11025 Hz sine wave at -1 dBFS), but its mp3 will have no clipping (on a digital peak meter)...

why would 11,025 Hz give you ISPs...?
all of peaks are on sample points, last time i checked...

[Alexey Lukin]

Originally Posted by ProducerP

I was just trying to understand if such high levels of ISP on my wav were normal.

For highly crushed music it is not unusual to have a few peaks per track that exceed +2 dB.

Originally Posted by stinkyfingers

why would 11,025 Hz give you ISPs...?
all of peaks are on sample points, last time i checked...

Try the initial phase of 45°

[stinkyfingers]

^ but you shift the phase to do that...
that is not a "typical" 11,025 Hz sine produced @ 44.1k...

[Alexey Lukin]

Why not? There's no preference for a particular phase of the waveform as you generate or sample it through A/D.

[Lagerfeldt]

Originally Posted by ProducerP

I just don't know if it's ok to leave oversampling off, because with the hard limiting that I'm doing, ISPs easily go above +1.5dB... Not a problem for the wav, but I don't know for the mp3. To me it sounds good... but do you think it's too much?

If you're making music just for you and your system, perhaps. However, if you're making music that will be played through a variety of systems and processing (such as streaming or other lossy encoding/processing) you should avoid ISPs. They can cause distortion and pumping during playback on systems with internal speakers and built-in protection, such as laptops and iMacs. It will also cause problems on Spotify with the always-on limiter.

Clipped signals in broadcast processing is also problematic as it confuses the cascaded phase rotators. It doesn't sound louder or better, only more distorted on air.

Also, you have no way of knowing how a random DA will handle the headroom being exceeded - you don't know how audible the problem will be. It's an unknown and unnecessary factor in my opinion.

If you're not doing it for the loudness, but you like the sound of the clipping or hard limiting, then at least lower the output ceiling so ISPs won't cause constant 0 dBFS+ signals.

[stinkyfingers]

Originally Posted by Alexey Lukin

Why not? There's no preference for a particular phase of the waveform as you generate or sample it through A/D.

ah, ok...i didn't realize you were using analog (conversion) for that example...
i thought you meant all ITB...
my head is stuck ITB... :P

[Elvis Gotta Gun]

Originally Posted by ProducerP

Hmm damn. I bounced a wav with a ceiling of -0.30 dBFS and I loaded it into the offline TT Dynamic Range meter, which gives me overall dynamic range, rms and peak of a track. It says that the peak is "over", meaning that i still got overs even with a ceiling of -0.30 dBFS. How can you be sure of that exact "-0.19 dBFS"?

If you have zero lookahead and generous attack settings, you can clip Pro-L easily when pushing high levels into it.

To echo what Lagerfeldt already suggested, maybe experiment with different parameters.

[haryy]

Originally Posted by Alexey Lukin

The artifacts of mp3 are usually much worse than artifacts because of ISPs. When mastering for mp3, I would concentrate on mp3 clipping and keep it in control, not ISPs.

Don't hunt for every instance of clipping: not every clipping is audible. Make sure that you are not clipping badly.

Yep, that's what i think, too.