Airwindows ToTape7: Free Mac/Windows/Linux/Pi CLAP/AU/VST3/VST2/LV2/Rack

[WackyZoundz]

In fact that sounds like a quality decrease..

FIR filters just work differently than IIR filters, they are not better per se. It's like comparing a standard delay to a reverse delay. The latter takes more RAM without providing higher quality, it just works in a different way (Larger buffer, obviously).

Why would you need steeper than 2pole filters for AA if the samplerate is so high?

Because the cutoff frequency (depending on genre, style etc) has to start somewhere between 16 kHz and 22.05 kHz with a slope that reaches -96 dBFS at 22.05 kHz (in case you aim at Audio CDs or standard streaming formats). Do the math, 2nd order doesn't get you anywhere. You will be much closer to a order of 100.

No, oversampling is Upsample + Very steep filter + downsample.

If you use a samplerate of 44.1 kHz you already oversample (1x) because that's more than twice of 20 kHz (what you're working with). You use at least two samples for each frequency. You can test what happens if you don't oversample at all by using a samplerate reducer (Effect plugin) and setting its samplerate to 22.05 kHz. Then you can hear aliasing wreaking havoc (if you don't use a lowpass filter which is also a part of your AD/DA converters by the way).

If you use say 3 plugins with oversampling that is 6 times the signal gets resampled and filtered, vs none simply using a high samplerate from the start.

In reality you use a lot of plugins and most of them don't need (high) oversampling. You only need it when using aggressive waveshaping (Overtones!) or to reduce filter warping when using IIR filters set to higher frequencies (Usually starting at a tenth of the samplerate and getting exponentially worse the closer it gets to the Nyquist frequency). But a highpass filter set to 20-40 Hz does not benefit from high oversampling nor does a compressor with slow attack and release times ("Slow" as in "Does not distort the signal to a degree it generates additional overtones") nor does a gain fader nor a chorus/flanger/phaser nor a goniometer nor a simple rompler. It's just a waste of CPU and RAM.

In case I use plugins in series which require high oversampling (which is usually the case) I put them together in a container (Modular plugin host that acts like a plugin) which is then loaded in a wrapper which oversamples the entire container. In rare cases I end up with a maximum of four wrapped containers in series with each adding either 12 or 18 samples of latency (Double or quadruple oversampling, 88.4/176.4 kHz) which is still low enough for live performance. In most cases I'm fine with just one or two containers in a daisy chain (from input to master stage). I get a good sound for a fraction of the usual CPU/RAM cost, don't have to bounce and downsample later (No offline rendering issues), I can record "live" out of the DAW at any time as if it was an analog mixer and already have a finished track for mastering. And I can also use plugins which don't support high samplerates. It's unusual but it works.

[EfreetiSultan]

In fact that sounds like a quality decrease..

FIR filters just work differently than IIR filters, they are not better per se. It's like comparing a standard delay to a reverse delay. The latter takes more RAM without providing higher quality, it just works in a different way (Larger buffer, obviously).

Why would you need steeper than 2pole filters for AA if the samplerate is so high?

Because the cutoff frequency (depending on genre, style etc) has to start somewhere between 16 kHz and 22.05 kHz with a slope that reaches -96 dBFS at 22.05 kHz (in case you aim at Audio CDs or standard streaming formats). Do the math, 2nd order doesn't get you anywhere. You will be much closer to a order of 100.

No, oversampling is Upsample + Very steep filter + downsample.

If you use a samplerate of 44.1 kHz you already oversample (1x) because that's more than twice of 20 kHz (what you're working with). You use at least two samples for each frequency. You can test what happens if you don't oversample at all by using a samplerate reducer (Effect plugin) and setting its samplerate to 22.05 kHz. Then you can hear aliasing wreaking havoc (if you don't use a lowpass filter which is also a part of your AD/DA converters by the way).

If you use say 3 plugins with oversampling that is 6 times the signal gets resampled and filtered, vs none simply using a high samplerate from the start.

In reality you use a lot of plugins and most of them don't need (high) oversampling. You only need it when using aggressive waveshaping (Overtones!) or to reduce filter warping when using IIR filters set to higher frequencies (Usually starting at a tenth of the samplerate and getting exponentially worse the closer it gets to the Nyquist frequency). But a highpass filter set to 20-40 Hz does not benefit from high oversampling nor does a compressor with slow attack and release times ("Slow" as in "Does not distort the signal to a degree it generates additional overtones") nor does a gain fader nor a chorus/flanger/phaser nor a goniometer nor a simple rompler. It's just a waste of CPU and RAM.

In case I use plugins in series which require high oversampling (which is usually the case) I put them together in a container (Modular plugin host that acts like a plugin) which is then loaded in a wrapper which oversamples the entire container. In rare cases I end up with a maximum of four wrapped containers in series with each adding either 12 or 18 samples of latency (Double or quadruple oversampling, 88.4/176.4 kHz) which is still low enough for live performance. In most cases I'm fine with just one or two containers in a daisy chain (from input to master stage). I get a good sound for a fraction of the usual CPU/RAM cost, don't have to bounce and downsample later (No offline rendering issues), I can record "live" out of the DAW at any time as if it was an analog mixer and already have a finished track for mastering. And I can also use plugins which don't support high samplerates. It's unusual but it works.

why do the cutoff frequency for the (manual) AA filter have to be between 16 and 22khz, if youre running in a samplerate of say 96khz? couldnt you just as well have a "relaxed" 12db slope filter at say 24khz?
I realise that in the very end, as the last step of mastering, is to downsample to 44.1khz - and in that process another AA filter is ofc added as part of the SRC. Im referring to the manual bandlimit/ lp filter you would add to your busses in the mixing process at a high samplerate. THAT specific filter could be very "relaxed", no?

I really like your method of using a wrapper to oversample an entire effect chain. That seems like the best of both worlds. Its given me something to ponder.

the rest, in general, i agree with - and/or have nothing to add to.

[jinxtigr]

Don't get too fussed with all this. There is another version coming, not this weekend but probably the one after that.

The Dubly stuff is getting fixed HARDCORE and ToTape7 will end up more of a curiosity. It's the Dubly2 that is gritting everything up and there's a Dubly3 that is strikingly different and, I think, a lot more fun to get sounds with, and a lot better at it. If you want to get hype, don't get hype for ToTape7, get hype for ToTape8 because you're not going to have to wait another two years for it.

[FabienTDR]

I love KVR because the people here are nice and full of audio wisdom (actually not his Quote).

Would love to hear some thoughts on this from Fabien, if he wants to chime in? As we all know TDR - Plugins have their very own unique way of dealing with Aliasing and the use of Oversampling, or just Sampling in this case. The use of their Filter - Bundle is also very related to this Topic. viewtopic.php?t=603998

lol about the creative quoting

I don't want to derail Chris' product thread, and honestly don't really care what other devs decide or how they invest their resources. We like our offers to be antialiased by default, and the habit offers us a distinct competitive advantage in certain circles, but carrying this responsibility also asks for significant development effort. Hence the limited product range.

Chris on the other hand has a far greater creative output, he can experiment a lot. I perfectly understand his approach.

I personally see implicit antialiasing as the only proper way to handle this wrong math, this open violation of the sampling theorem, this historical error. Whenever a plugin decides to introduce nonlinearity, it should also offer reasonable ways to prevent aliasing - or at least properly document the erroneous behaviour, even if creatively useful. But that's still niche it seems.

Chris tells his audience to use very high samplerates instead. That's a reasonable and quick way to push away the problem (at the price of not offering the best possible antialiasing effectiveness and efficiency to his audience). It would otherwise be difficult to justify the development effort in his freeware focus. We started to sell software to justify that huge effort, and due to it, also can't experiment much in public like Chris does. No free lunches.