Aliasing in softsynths

[tomdot]

I know how to test for digital artifacts in plugins (in short, run sine sweeps through the plugin and look at the frequency spectrum).

However, this involves setting up a repeatable method to put some gain through the plugin for comparison’s sake.

How can I do this for softsynths too? I figure that the ideal case is to adjust the synth to output a sine wave, then play a note while I conduct a sweep.

Is there a better way? I can see how time consuming my approach would be, so am hoping for a quicker method. Of course, I can listen for the wishy washy aliasing sound as I run through preset patches I suppose, so is that a viable alternative?

[Martin Walker]

I've done loads of thorough testing of aliasing in plug-in effects, but have never (yet) attempted to do the same with a softsynth, because...

First of all, I suspect this will depend on whether the waveforms are from a set of samples, or calculated/generated mathematically.

Second, a lot will depend on whether the waveforms are band-limited, and how many of them are spread across the keyboard range (having lots of samples, each only played back across a few semitones, is generally going to sound cleaner than relying on fewer samples and much wider transposition). Other sources of aliasing within the synth will of course include saturation, 'analogue' EQs/filters and so on.

So it don't think it's going to be fair to just come up with a sinewave from a synth and then sweep it up and down over a large range, as I suspect most synths will be using all sorts of 'clever tricks'/algorithms to minimise any remaining aliasing with practical use.

To be honest, I'd be inclined to play single notes over the whole keyboard range from a variety of presets and see if you can hear anything untoward (particularly towards the higher limits of the keyboard and any octave shifts within the preset), and this WILL take a long time. I'd also be inclined to play chords as well, to listen out for any intermodulation distortion. Temporarily disabling any internal FX can also help expose 'nasties' that can end up covered by a quick splash of chorus/reverb etc (but then again, if it's been covered up successfully in a finished preset using such means, does it matter?)

One thing you might like to try, and that's the 'quality' setting offered by many softsynths nowadays. Set this to its lowest 'eco' setting and listen out for harshness at the top end, and then switch to higher quality settings to see if you can hear any improvement. That at least will help you choose the most sensible setting for normal day to day use, but once again, some softsynths offer to always render the final track with 'obscene' quality settings.

For me, the big difference between softsynths and software plug-ins is that with plug-ins, you at least know the sorts of sounds you're likely to be putting through them, and the associated frequency response/transient content and so on, so you can amass a core of test files that quickly expose particular artefacts (drum/percussion sounds will for instance help you hear the effects of compressors/saturation/preamps in very different ways than when feeding in sustained sounds).

The problem with softsynths is a the huge variety of potential sounds that may make up a single preset, and while the factory presets may use conservative transpose/octave settings, other sound designers may 'push' the engine in more extreme directions than the developers envisioned resulting in harshness that might offend some users, but delight others.

Good luck!

Martin

[tomdot]

Thanks for the thought out reply, and it’s confirming what I already had figured out: that sounds will vary so much it would be impossible to. Carry out a fair test.

I feel as though playing around with notes and chords across the keys will tell me most of what I need to know anyway.

[RichardT]

A lot of soft synths churn out vast amounts of high frequency sound - sometimes up to 40kHz! It’s well worth looking at that and sometimes apply filters to get rid of it.

[muzines]

A lot of soft synths churn out vast amounts of high frequency sound - sometimes up to 40kHz!

Really? Which ones? They should be bandlimited, and not be producing anything approaching Nyquist at typical 44/48KHz sample rates, ie not much about 20KHz.

Or do you mean when running them at eg 96KHz?

[RichardT]

A lot of soft synths churn out vast amounts of high frequency sound - sometimes up to 40kHz!

Really? Which ones? They should be bandlimited, and not be producing anything approaching Nyquist at typical 44/48KHz sample rates, ie not much about 20KHz.

Or do you mean when running them at eg 96KHz?

Yes, running at 96kHz.

[Martin Walker]

Phew - that's a relief then

Otherwise, the aliasing would sound horrendous

[Hugh Robjohns]

A lot of soft synths churn out vast amounts of high frequency sound - sometimes up to 40kHz! It’s well worth looking at that and sometimes apply filters to get rid of it.

Er... so do analogue synths. a squarewave, for example generates a huge train of harmonics that theoretically go on forever (albeit at diminishing levels).

The first few harmonics that make up a 4kHz square wave, for example, are 12kHz, 20kHz, 28kHz, 36kHz and 44kHz. Without them (and many more -- most analogue square waves are made with at least 10 odd harmonics) it wouldn't be a square wave.

So, if your system is running at 96kHz I would expect to see all of those harmonics running up to 40kHz or so. That's not aliasing, it's the way it should be. After all, the only reason to be running at 96kHz in the first place is to create a wider bandwidth than is available with lower sample rates.

Obviously, if running at a base sample rate, anti-alias filtering filtering should be in place to restrict the bandwidth to around 20kHz.

[N i g e l]

I find aliasing most detectable in the upper octaves by bending the pitch.
When pitch bending going up in frequency, the aliasing is comming down in frequency and isnt harmonic.

[Martin Walker]

I find aliasing most detectable in the upper octaves by bending the pitch.
When pitch bending going up in frequency, the aliasing is comming down in frequency and isnt harmonic.

Great approach N i g e l - this is exactly the same as using a swept sinewave to test a plug-in. And of course many (if not most) softsynths let you choose pitchbend range, typically in semitone steps up to an octave, which would be quite sufficient to hear any aliasing.

[Folderol]

A plugin must know the sample rate it's being driven at, so surely it should make the necessary adjustments itself, even if that's just a 2nd order low-pass filter on the output.

[Martin Walker]

Indeed, except that there are many opportunities earlier in the coded signal chain to generate frequencies way beyond half the sample rate that can result in nasty aliasing, unless the results are band-limited at each stage.

[tomdot]

I like the pitch bending idea. That makes a lot of sense!

In people’s opinions, and purely from the viewpoint that aliasing is a digital phenomenon that should exist in analog hardware, what softsynths do the best in this regard?

[SafeandSound Mastering]

I imagine you meant should not exist in analogue there, 2 synths that do really well for me are u-He Repro-1 (that synth planned to cook my old 2600 i7 CPU in HQ mode last time I had it in a track so I kept HQ off and it still sounded great to me.) and Diva. It is not usually a major issue in a mix unless there is something quite wrong and that synth is loud and doing soemthjign particularly challenging. Dune 3 is a smooth sounding synth I don't recall anything sounding like aliasing from that.. it is not always a favourite though.

u-He do well.. I only like Repro, Diva and also Bazille oh and Zebra II can be ok for some sounds, hold on that's most of their synths ! TAL-U-No-X is great as well although the Juno /-106 is not strictly analogue. I think it has digital oscillators and an analogue filter/chorus.

I don't generally hear aliasing from those synths. I can hear aliasing when using the filter on my Virus Snow.. though that is a great sounding synth and part of the punchy slightly aggressive sound could actually be, in part aliasing. I am fairly sure I heard some unpleasantness from Vanguard2 when using very open filter settings with high resonance which is unsurprising as it was released initially in 2004. Was that aliasing it is dificult to say I was not really in technical mode with a analyser on it at the time.

Synths are funny things to me (totally amazing though).. they jump out at me with strengths so I have rather a lot of them and tend to reach for specific ones for a specific sound type.. VA/FM/WT/Additive/Virus specific (Formant complex) etc. Though they sometimes surprise you and fit right in when not expected. I don't really consider aliasing much and tend to put it down to synth character.. I think we need a bit of all sorts for music.

The worst aliasing I have ever heard from any audio device was a very well known compressor shipped within a famous DAW. It was one of the worst compressors I have ever used. It used to literally repeatably squeak with aliasing (assumed) at times. I won't name names !

[muzines]

I won't name names !

Name and shame for aliasing crimes!

(Does it still behave like this, or was it fixed?)

[tomdot]

I hear about u-He a lot, and I have the CM versions of Bazille and Zebra. Bazille sounds really good.

I feel you’re talking about the old Logic Compressor, which aliases a lot. The new Platinum one is fantastic, but only if you use the actual Platinum algo, or the (I think) Studio VCA. No aliasing there, but creating a 1176-type compressor that doesn’t alias is super tough.

[SafeandSound Mastering]

I am not sure as I have not used the software of concern for quite a while.