I used the same image, the Spectrogram is paused and I’ve just changed the settings for demonstration purposes, so yes, mine show aliasing on both, but yours don’t…
Just do the test again with x8 OS and adjust the settings to display only audible range, you’ll see that there will be no aliasing at all, the “vertical lines” are only FFT noise and harmonics.
1 Like
Giancarlo Del Sordo from Acustica Audio is sending to you a message:
"All software clippers in the world MUST create aliasing in the absence of oversampling, by the very nature of the object.
Fire The Clip has two different oversampling systems, and they are absolutely necessary to reject aliasing.
The first is an oversampling called “low quality,” it is actually a minimum phase IIR type and therefore goes to impact the phase. We have seen that some oversampling systems in other products are limited to this, for a power consumption argument. The reject of this filter, however, is excellent.
We have provided a “high quality” fir oversampling, linear phase, so with “perfect” phase and frequency response and slightly worse reject. The idea of finding a compromise with the reject comes from the choice of impulse response, because we decided to favor transient response (in any case, oversampling filter design is always a compromise). However, the same result is achieved by resorting to higher oversampling values.
In general, the oversampling system of Fire The Clip is superior to what could be achieved by the use of stock library functions such as those contained in Juce, because of a number of rather driven optimizations that are completely absent in other libraries. In fact it is a DFT implementation of a polyphase filter in which we have retained some taps in direct convolution to limit numerical errors."
i think you don´t know how to use the tools, are you a troll or something?
2 Likes
Comparing it to juce stock library is hilarious argument. But it is not really relevant what they claim or how they argument for their design. The measurement shows the outcome. And if they use juce, they should stop doing that since it is not working according to VST3-SDK guide lines it it breaks work flow in cubase. Ok, maybe I should not bashing it to much, it is a free-bee after all. And by the way, they classify their plugin as Dynamics where it should be Distortion.
anyway, before i got it for free, I demoed it against the black salt audio clipper. went with the bsa one, sounded better and dead simple to use.
No. They are not. You can see this with static signal generator and the spectrum analyser. It is not as visual and you need to track the non aliasing vs aliasing artefacts manually.
With this you can also measure the rejection. For a 19.9K signal the rejection is about 40 dB at 8x. It does not make sense to me, It is about 6 dB per octave.
But the sweep does that even when you don’t insert any plugin ! No sine wave has spot on frequency, it always produces noise in the rest of the spectrum, plus when you render a sweep it creates random fluctuation in this noise.
What you see on the spectrogram is all this noise that fluctuates, because you simply put the sensitivity too high ! Isn’t it that obvious for you that if you set it to display -240 dB it will indeed show that noise ? What’s that, seriously.
No human being is able to detect noise below -80 or -90 dB when the device is set to a normal listening level. The ambient noise will always be louder, even in a well treated studio. Here we are talking about noise that is below -130 dB, this is almost the noise floor of a 24 bit file !
The thing is that SuperVision’s spectrogram module has a quite low quality display, and this makes the lines more or less obvious depending on the FFT size, so depending on the settings, you will or will not see those lines.
Also note that with low FFT size the noise may appear louder than it is, so you may want to set it to the highest value for more precision about the noise level.
In the following picture the Spectrogram FFT is set to 1024 because for some reason the vertical lines are more were more obvious with that value, and Spectum Curve is set to Multi.
Notice the rumble across all the spectrum (on the spectrum curve) when the sweep is occurring.
Not only that, but this noise has a lot of movement when the sweep is occurring, and that’s what produces the vertical lines on the spectrogram.
Now here’s what the spectrum curve looks like with a constant frequency. There’s a low freq rumble and a lot of low level harmonics above the fundamental. Harmonics are also added below the fundamental as it increases in frequency.
The following GIF shows a constant frequency at around 2 kHz, and we can still see that the level of the noise oscillates. Just saying that the noise level at this frequency is around -180 dB !
I just want to note that I have tested it with other synth plugins and they don’t have any low freq rumble, although there are harmonics and the noise still oscillates a bit.
So is that a defect of the TestGenerator plugin ?
Nevertheless, with all this information, you can now know that the “vertical lines” aren’t aliasing, and the noise you see is way below the 24 bit noise floor anyway…
1 Like
First of all. Yes, test generator build in in cubase is far from perfect. I use EMO from waves for sine, and Izotope RX for sweep. And yes, sweep is a modulated signal so it will have some “artefacts”.
This is EMO without the plugin.
Same signal through the plugin at various oversampling.
No oversampling:
It have a lot of harmonics below the test tone. This might be the function of the plugin, or it might be a unwanted artefact. If it was intended to be there is should NOT be removed with oversampling.
128x oversampling:
Starting to look good. But it uses insane amount of CPU.
The one with no oversampling is supposed to have limitation on it’s aliasing.
So this is 4x.
4x oversampling:
Worst anti aliasing filter I have ever seen.
Giancarlo:
"the oversampling problem of stock libraries, on the other hand, is a major problem, since the clipper is highly dependent on its quality. We tried that route as well and did not like the end result. When a product is very successful on the various platforms in the end there is always a practical reason behind it: it is undeniable that Fire The Clip has had some media success, even when it cost more than other solutions and people already had a definitely good clipper among their purchased tools, such as standard clip.
As for classification, for us Fire The Clip is a dynamic loudness control tool to be used mainly in mastering and not a distortion tool. It’s true that it can be used as a saturator for creative effects and obvious distortion effects, but lately the main purpose for clipping tools with good oversampling is about dynamics control, which is why Fire has the True Peak feature. For those who don’t know, True Peak is always achieved by some sort of limiter-the clip alone is insufficient to solve the problem of intersample peaks"
@cubace At 4x it’s normal that is may still present aliasing. It just depends on the nature of the plugin and how hard you push it. If the generated harmonics are extremely loud, they will eventually reach the nyquist, and at 48 kHz with 4x oversampling it is only 96 kHz…
Anyway, if you find any aliasing at such OS, it will be far below the digital range of your final file.
The vertical lines on your spectrogram were just the artifacts from the sweep. If you do the spectrogram test again with a fixed frequency at 20 kHz, I highly doubt it will show any aliasing at 8x OS and above.
As a dynamic processing tool it is the most aliasing tool ever seen. A distortion sort of can get away with terrible handling, it is there to distort.
" It’s true that it can be used as a saturator for creative effects and obvious distortion effects, but lately the main purpose for clipping tools with good oversampling is about dynamics control, which is why Fire has the True Peak feature. "
It might have a True Peek feature, and it for sure can do oversampling, but it lacks filter that remove aliasing artefacts created with it’s distortion.
Are you a proxy for the company?
All other plugins that I have tested with oversampling removes much of the aliasing artefacts with 2x. With 4x it is very hard to see any traces. This one it is really hard to see that there is any difference at all with no oversampling and 4x oversampling. It seems like they use the oversampling for inter-sampling true peaks only, and not at all for removing aliasing artefacts.
What’s the relation between true peak and oversampling ?
True peak is only a matter of level, two adjacent samples will see their level reduced so that the reconstructed intersample signal doesn’t go above the threshold, while oversampling is for increasing the frequency range in order to reduce aliasing… Totally different things.
If you are going to find a true peak, you need to find the peaks that are between sample points. This is done with extrapolation, that is essential a oversampling.
This is done internally, that doesn’t even affect the oversample setting, you can still enable true peak and have OS disabled…
Well, that is a guess. Since they do not do have much of aliasing filtering and they seems to be happy with their true peek things. Oversampling is also done internally.
These are facts, those are two separate process, it isn’t the plugin’s oversampling setting that controls how the true peak detection behaves.
Now I end this discussion here.
2 Likes
Well, I dont have access to code, so my analyse is based on measurement. They are in previous posts, if thing they are wrong you are more than welcome to point out what is wrong with them.
Don’t feed the troll. Discussions with this guy are pointless anyway. Thanks for your contribution and your effort! @Louis_R
PS: In the end aliasing is mostly not even an issue at all, as it is easily below the hearing threshold, when using OS in most plugins. Its way more important how a plugin sounds, not how it measures. Best example are KUSH compressor plugins. A lot of them have quite some aliasing going on, but they often still sound better for certain applications then other compressors.
1 Like
Aliasing is a major reason why pros that can avoid plugins uses analog gear. If you and/or your clients dont care or don’t hear it you should of course pick what you like.
KUSH is a better choice. It also need oversampling, and they have it and it works, even if their distortion types should benefit from more than 2x. And the reason to measure aliasing is that it sounds bad. It does not automatic sound good if does not aliasing, but i sounds more consistent and work in the same way on all projects. If it aliasing it will be very different on a 44.1, 48 or 96 khz project.