goes crazy in k meter but in waves it is still over at 0.2 red

i guess i am lucky i have one day left on my waves demo as that is the one i trust the most

k meter is like 0.8 or so over.. much more.

Toneboosters shows the tru peak as clipped 0.3

sorry mate it is in finding with my other reports so far. It is not doing any better at isp than any other limiter

and toneboosters barricade is the best at isp of anything on the market. (including pro-l)

you know I never even tried toneboosters limiter im using the demo of there meter the thing with me is that I never push my limiter at all. Most of the time it does not even hit the limiter and when it does its like max at 1 db reduction. also I am working in 32bit 48 hmm maybe at diforent bits depths it works different?

in this case i am just going by the test file.

since my music is destined for cd and itunes i see no value working in anything other than 24/41

very very true but I noticed this thing with stuff like amp sims and recording classical instruments that at higher quality settings the tone is much fuller and alot more balanced. try it out record a track using any amp sim like guitar rig 16 44.1 then at 32 or 24, 98k there is a huge difference and when its rendered down back to 16 44. it preserves most of the quality

I didn't check it yet because I don't believe in this For me, the should sound the same but I dint't make null tests yet.



So, either _all_ of those meters are wrong or we use different methods for ISP calculations.

My approach is to use pure ideal mathematical approximation of audio signal. All DSP books say that analog audio signal is sinc function interpolation where _all_ digital samples affect audio signal in given point. Of course, in practial usage the influence of samples is restricted with large enough window. And DACs give you only similarity to this true mathematical audio signal.

What I can suggest:

1. My audio sample file is very short (~5600 non-zero samples) so I it can be mathematically proven what the ideal peak value for this sample file.
2. You can convert this sample audio file with the best SRC you have from 44.1 to 192 kHz (4.35x oversampling) and then look at regular peak values. EBU R 128 declares that 4x oversampling can be used for true peak metering (at the same time it says that such meters can underread 0.5 dB of signal value).
3. The best way is to plug DAC to digital oscilloscope then record true signal value at very high sample rate and then analyze it.

My conclusions:

For example, EBU R 128 says about 4x oversampling. But it doesn't declare what oversampling algorithm should be used. And I think most plugins use the _simpliest_ algorithm for it. Just to decrease CPU usage. But for example polynomial oversampling doesn't give something like what you have after DAC conversion.

Sorry, but for mp3 I don't have information about peaks calculation. My old rule was to limit peaks to -0.5 dBFS before mp3 conversion.

My ISP metering uses 4x oversampling sinc function interpolation with window size of 256 samples. And then polynomical approximation by 4 points to find real signal maximum within these 4 oversampled points.

If you point me where I'm wrong, it should be good because in this case I can fix this plugin

Well, I've got pretty cool idea how to check ISP stuff. It can be done with simple impulse audio files. Let's start!

TEST#1

Wave file looks like this (where numbers are sample values, 0 is -inf dB and 1 is 0 dBFS):

... 0 0 0 0 +1 +1 0 0 0 0 ...

http://dl.dropbox.com/u/18475891/impulse_1.wav

Mathematically, maximum ISP value will be:

sin(pi/2) / (pi/2) * 2 = 4/pi = 2.098 dB

TB EBU (I downloaded latest demo from toneboosters) shows: 1.9 dB

L6 ISP detector: 1.275 (sample value) = 2.111 dB

As you can see TB EBU have 0.1 dB precision and L6 have 0.02 dB precision for this test. For L6 this value is hidden inside ISP limiting module. But I can just output it somewhere on the screen or I can move this detector out and compile simple true peak metering plugin. After that it can be interesting to create dedicated thread on this forum and analyze all known true peak meters!

TEST#2

... 0 0 0 0 +1 -1 0 0 0 0 ...

http://dl.dropbox.com/u/18475891/impulse_2.wav

ISP value calculated by mathematical numeric methods: 0.765 dB

TB EBU: 0.4 dB

L6 ISP: 0.799 dB

As you can see in this case L6 gives 0.03 dB error and TB EBU 0.3 dB.

TEST#3

strange file from my archives:

http://dl.dropbox.com/u/18475891/k_meter_bug.wav

TB EBU: -0.0 dB

L6 ISP: 1.001541 = 0.013 dB

And yes, ttoz was right! This file has ISP clipping inside! (Before, I told about this file by memory and seems I already forgot all details about this file, it is dated by october of 2011 so you can understand me)

The clipping is between samples 44110 and 44111 (right channel)!

But now we can apply L6 ISP limiting with ceiling -0.3 dBFS to this file to defeat this ISP clip!

After L6 ISP limit -0.3 dB: my ISP metering module shows -0.299637 dB

TB EBU: -0.3 dB

So, in this time I trust TB EBU!

Conclusion. Soon you, guys will have true ISP meter for free

Sorry for my ignorance...but what does "ISP" means ?

Inter Sample Peaks

Thanks I feel less stupid now !

Essentially, Inter Sample Peaks appear in all audio when it's passing through the digital to audio converter, since DAC tries to recreate the true analog waveform. The best way to check for inter sample peaks of *your particual DAC*, since they do it differently, would be to loopback its output back to the DAW and lower the volume a few dB so you would be able to see the peaks. It is not possible to completely remove ISPs, because of the DAC.

Therefore it's best not to make such loud masters since it ruins people's ears, and it sounds awful. In my opinion K-14 is the best standard there is. Average RMS of -14dB with maximum RMS peaks -10dB, 300ms integration time like VU meter, I'm old school Even then, ISPs can occur, but there's no comparison to these awful recordings ignorant people are making today, with 2-3dB of dynamics, in comparison to K-14 that gives you at least 10dB of dynamics. K-14 "zero" is at -14dB RMS, and peaks are at -10dB RMS. You still get the audio peaking nicely to digital 0 [0 dBFS] on occasions, if you mixed the track properly. It is also a good standard IMO for mixing, to show you if you overcompressed it or undercompressed it [if that's possible ].

I hope this explains it all?

Cheers!

As you indicated, EBU R128 does not specify the anti-aliasing filter, or any requirements for such filter except for a minimum of 4x oversampling, and a couple of optional elements (such as DC blocking, and high-frequency under read correction). Therefore differences can be expected for different implementations/products that measure true peaks, while these different implementations can all still be fully compliant with the standard.

I think the R128 (and ITU-R bs.1770) standard is not aiming for the mathematically exact true-peak estimate. Instead, it is aiming for clipping prevention in practical D/A converters and subsequent analog signal paths. You will have a hard time to find a D/A converter that produces true peak values that are equal to your mathematical true peak values, especially given the fact that in your -1/+1 signal sequences, virtually all signal energy is exactly at the Nyquist frequency. Practical D/A converters will always have some attenuation at Nyquist, and hence their output true peak will be lower than the mathematical (or exact) true peak value.

The -1 dB (FS) maximum true peak value can be interpreted as a safety margin for differences in true-peak estimation algorithms. As long as the deviation between estimated true peak and mathematically-exact true peak is less than 1.0 dB, one wouldn't expect analog clipping if the true-peak estimate is kept below -1.0 dBFS.

YES!...
i found that Voxengo Elephant is the most "accurate" limiter with ISPs that i have tested...
("better" than toneboosters, nugen, vladg...)
it would be interesting to see the whole lot with the same material...

Yeah! Now my version of ISP peak meter is available here:

http://vladgsound.wordpress.com/2012/07/15/simple-isp-peak-m eter-and-few-tests/

Who want to be a new thread starter?

I've got a plan how to compare true peak meters: I have test impulse files (0 0 +1 +1 0 0: impulse_1.wav) and (0 0 +1 -1 0 0: impulse_2.wav) and I can mathematically calculate true signal peaks for this files (see my link for details). Also I can add a couple more test impulse files (+1 -1 +1, +1 +1 -1 and maybe +1 +1 +1 for example) with calculated peak values. So we just check all known true peak meters agains these files and write results.

But how do we check ISP limiters? It must be done on real sound material, I suppose? For example, we limit the same file with -0.3 dBFS threshold with 5 different ISP limiters and then we analyze output with 5 different ISP meters and output results. Any ideas?




EDIT:
I don't agree about frequency response. Impulse: 0 +1 0 has frequency response of horizontal line (like white noise). Impulse: 0 +1 +1 0 has fall at Nyquist (-3 dB at ~11 Khz). But for impulse: 0 +1 -1 0 your're right. It's all high frequencies.

In this case we just can't easy compare different ISP meters. The only way is to attach digital oscilloscope to DAC and watch for clipping. Is it?

EDIT2:
We can compare all known ISP limiters against all known ISP meters without using mathematical calculations. Just limit one signal with all of them and write results of all meters. We can't say what limiter works right because we can't say what meter is right.

i can get fairly accurate readings for true peak with an oscilloscope (w/ 4x os)...
also, you can use iZotope Rx and zoom in to see the "analog" peak level of a waveform...
i'm busy today, but maybe tomorrow i can share some of my findings to compare...

I suddenly realized that I can use second audio interface at high sample rate (192 kHz) to get output from 1st audio interface at regular sample rate (44.1). And it'll give some info what are the true peak values after DAC without oscilloscope.

Also I suppose to avoid clipping in DAC test impulse should be +0.5 -0.5 but not +1 -1.



Thanks a lot! Waiting for your results...