request: alias free sample engine


(ffx) #1

These guys kindly now added Renoise 3.01 antialiasing capability:

http://src.infinitewave.ca

You can see that Renoise 3.01 actually processes samples in hq mode less good than Renoise 2.8.

Since Ableton 9.1+, Logic 8+ and even Core Audio since 10.6 resampling is nearly alias free, I wonder why the devs do not looking for such a modern alias free algorithm.

Is it possible? Well, IMO you really can hear the difference between Ableton 9.1 and Renoise 3. Ableton sounds smoother.


List of feature suggestions for Renoise
(joule) #2

+1, I think.

My impression is that I could hear a lot of aliasing on chip samples in Renoise compared to some other softwares.


(RANSOM) #3

guys have you tried to choose SINC interpolation for samples???

and that test didnt mentioned this thing either, isnt it?


(Zer0 Fly) #4

I’m bugged like hell about the aliasing, too. It prevents being able to do high pitched notes from samples, being single cycle chipstuff to other recorded. It aliases badly, cubic and sinc make very little difference to my ears. Only upping the samplerate to something high (i.e. 96khz) makes a difference.

I guess the results there of “cubic” and “precise” - precise means sinc interpolation. I find it funny that most pulse responses (there are none for renoise tho) look like a pretty sinc pulse with preringing when samplers with little aliasing are shown. IDK whether applying such filters or methods would work out for realtime sampler action, preringing means added latency the better the filter the longer, and variable sample playback rates (modulated pitch) means such filter would have to adapt their curoff constantly resulting in mad computations.

Don’t forget that probably many of the “good results” present audio tracks, or static resampling libs, or whatever, and not samplers with freely variable pitch? Trotzdem it’d be good to have tracker/sampler playback without aliasing, i.e. prefiltering or whatever. Will chop cpu, but I think it has to be possible?

Are there actual “samplers” with pitch modulation with real alias clean results and low cpu usage? I think there have to be methods to achieve this. Renoise makes o.k. tracker style sound, but has no option for real clean stuff - it’s a pity. And it won’t even provide methods for at least prefiltering stuff with autospread over the octaves, this can reduce aliasing as long as no hefty pitch modulations are used.


(Akiz) #5

I think that Highlife has a great engine.


(ffx) #6

Hm ableton 9.1+, core audio and logic 8+ obviously use algorithms that are 1. very efficient and 2. almost alias free. Alias free = warm sound. Renoise has “nice high frequencies”, but sounds quite digital from sample play. Maybe we expect too much and such a good algorithm is only available from some secret lab (like fraunhofer) that licenses it?

Don’t want to use a sampler in Renoise. That’s one of the main features of Renoise, so it should do it in the best way possible.

No, they tested offline sinc / hq and live mode (I sent it).


(Djeroek) #7

Isn’t there a difference between playback engine & when actually rendering a track to .wav?


(toblerpone) #8

A simple realtime and off-line render hq button would be nice, even if it’s plain old oversampling, remember, it’s not just the sampler that will alias over, it’s the entire signal path.


(Zer0 Fly) #9

Aliasing will affect the signal path from the point on where it’s generated - it is just passed on the whole chain. So if the sampler will generate aliasing, all fx after it will effect the aliasing too, making it worse in many situations. Samplers generate aliasing, some fx might too (waveshapers/distortions, extreme modulated delay lines as in flager/chorus). Aliasing can’t really be removed afterwards, you have to make sure each generator or fx simply won’t create aliasing. For a sampler this could include prefiltering the sample at its own rate so it won’t contain any frequencies above half the resampling target rate when transformed - trying to filter them out at the new rate after the resampling when destruction was already done makes no sense I guess. Oversampling and then using a halfband filter might help, but in extreme speedups of the original sample you’ll still have aliasing. As I said, with sound sweeps going very high I can easily trigger audible aliasing even at 96khz sampling rate.

I also think when aliasing is in the signal path, even if inaudible because reflected freqs are too high to percieve, it might fuck up further processing when certain extreme fx are used, i.e. make distortion modules sound not singing/harmonic/bright in the high freqs but cause noisy/harsh/inharmonic action robbing the sound of purity and brilliance bad time. I mean this traditional worship of analog stuff is not only because of nonlinear colouring of the sound, but also because it’s fucking warm and bright and clear sounding at the same time - it simply won’t alias if constructed right.

I still don’t know, with all those sine sweep spectrum plots with some having reflection and some seemingly being very clean, if those are realtime action plots or offline somek tap sinc filtering results that never could be played back in realtime? Because those sinc filters will rob some to many milliseconds latency when blocking real hard, and live actions should need below 10ms latency!


(muckleby) #10

if this gets renoise any closer to the sound i got from the EMu e64 im all in.

nicest soft sampler i’ve heard was a reaktor ensemble that used 13pt Sinc Interpolation realtime and 512pt offline, sounded mint but was too buggy to bare.

i would love something like this quality natively but can anyone recommend any similarly pristine vst samplers in the mean time?


(toblerpone) #11

trying to filter them out at the new rate after the resampling when destruction was already done makes no sense I guess

I never suggested that :stuck_out_tongue:

I guarantee you will have a tough time practicallyfinding an effect which will sound worse at a higher rate, of course if you already have it there you’re in a pity. I would be great if I could simply wrap around a dsp in a oversample device. Right now the only fix is ‘don’t use that dsp’ which doesn’t seem nice at all. Plugins normally support 192khz just fine and 192 is loads better than the 44 everyone is running, especially for most, not some, renoise dsps.

(for those unaware, higher than 96 can be achieved via asio control panel and via rewire by using a master daw where you can specify a rate higher than 96khz)

EDIT: you know what would be cool, a device like the oversampler above, but for ticks, we’d get our HFOs :smiley:


(ffx) #12

Djeroek: Yes, blue wrote about it here:https://forum.renoise.com/t/normalized-output-volume/43356


(danoise) #13

AFAIK the waveform editor does apply bandlimiting when doing sample rate conversions, but this is not done during playback/rendering.
Might explain the difference between 2.8 and 3.0, that the tests were done differently.

Would be nice if infinitewave were more open about how each test was set up?

I am curious not only about Renoise, but many of the other targets.


(muckleby) #14

after reading this thread i had a look for some advanced interpolation vsts and found shortcircuit. it has an option to use sinc in realtime and its impressive, a single string sample sounds musical and smooth across the whole keyboard, just like my old emu :slight_smile:
id really love to see this as a realtime option natively. with the cpu power we have today the resource hit is no biggie.


(TheBellows) #15

Who is this Alias guy and where do i get this free engine sample everyone is bragging about? :badteeth:


(danoise) #16

shortcircuit. it has an option to use sinc in realtime

Without pointing to the other complexities of interpolation, and why sinc is not always the “best possible” choice, I would simply like to point out that R3 has a per-sample interpolation option - including a 64 pointsinc mode :slight_smile:

Source:http://tutorials.renoise.com/wiki/Sampler#Options


(Zer0 Fly) #17

why not offer option to blow your butterworth 8n or “lr8” iir filter at the samples before interpolation with cutoff tuned relative to pitch to have maximum stopband all over target-above-nyquist-range & give a shit about that it’ll mess the phase around the cutoff 'n it’ll need windowing of the iir trail after note cut…? Better than nothing, I’d say.


(gentleclockdivider) #18

You gotta see the improvements of reason 6.5 to 7.0

Yep renoise could do better


(muckleby) #19

Without pointing to the other complexities of interpolation, and why sinc is not always the “best possible” choice, I would simply like to point out that R3 has a per-sample interpolation option - including a 64 pointsinc mode :slight_smile:

Source:http://tutorials.renoise.com/wiki/Sampler#Options

oops, howd i miss that!

agreed not always the best option but for strings and soft pads it’s mint.