The sound of aliasing?

[frejahel]

I am new at this forum and while reading up on various instruments, one obvious complaint of an instrument's quality is audible effects of aliasing. I have also seen people say that they notice these effects immediately.

Being somewhat of a beginner in the practical know-how department of sound design, mixing etc., I know how aliasing works in theory: i.e., going from continuous to discrete in the time or frequency domain will yield periodisation in the other. Sampling a continuous time signal, will give rise to a periodisation in the frequency domain (i.e., the continuous Discrete Time Fourier Transform) with a period of the sample frequency. When the signal is reconstructed, it will appear as though the frequency components above half the sample frequency are folded around Fs/2.

In other words, the reconstructed signal will now contain some unintentional frequency components if the sample rate is too low for the bandwidth of the signal. My question is: How do you recognise this by ear, given that aliasing is dependant on the character of the original signal and the sampling frequency? Will aliasing usually be recognised by its distorting the signal in unnatural ways and cause unpleasant digital artefacts for more "complex" (outside of school book examples with a few sine waves that is ) sounds?

Best regards,
Maria

[Music Engineer]

in this thread

http://www.kvraudio.com/forum/viewtopic.php?p=3298389

i posted an example for a sawtooth wave with and without aliasing.

[IrionDaRonin]

Interesting video about aliasing that i always recommend.
Anyway, it seems that you already know about this.

Have a nice day and welcome to KVR

[bbaggins]

What aliasing sounds like depends on the material being played, because it's derived directly from that material.

For example, with high plucked notes such as an acoustic guitar, it might sound like birds chirping. With a low percussive sound such as a kick drum, it might just make the initial hit sound like a cardboard box. With a steady-state tone it might sound like fast frequency modulation. Aliasing is weirdest on bent notes like an electric guitar or synth lead, because the note might be bending upward while the aliasing shifts downward.

One of the most obvious manifestations of aliasing is spoken word encoded at low bitrates. Listen closely to the voiceover in the video linked by IrionDaRonin above, which ironically has noticeable aliasing. It's most noticeable on "S"s and "T"s.

In all cases, aliasing sounds bad. Our ears are especially sensitive to it because the aliased frequencies are not harmonically related to the source tones, making them stand out. It's like intermodulation distortion in that respect - even a little bit is offensive to the ear.

[Sendy]

A good test to check for aliasing in synths is to play a very high note and apply a slow, deep vibrato. Start with a sawtooth wave and listen. Aliasing moves in the opposite direction to the desired signal, so if you hear (or see on a spectrograph) some components going up in pitch while others are going down, and no effects or other synth functions are engaged, you can reasonably bet that you've got aliasing.

Most competent modern synths should pass this test with flying colours. You can then try doing the same but with a PWM sound, with a sound using oscillator sync, FM, filter FM, etc. These types of output pose a tougher challenge for coders to get right.

A sort of related problem is when anti-aliasing measures rob the sound of some of it's power. Now you don't have so much aliasing but, for example, PWM might lose it's edge and sound watered down. When you hear a high pitched PWM lead done right it's a fantastically magestic sound, but for years softsynths have rendered them in a very washed-out way for fear of aliasing.

Another pitfall is that with certain modulations such as FM and audio-rate mods, it can be tough sometimes to work out if you're hearing aliasing or modulation effects, though in general if both the carrier and modulator are vibrato-ing together you should expect to hear only a rising and falling, not an accompanying contrary motion signal. (If you're just vibrato-ing for example the carrier, you'll probably hear a much more complex sound with moving sidebands etc).

Also, as a side-note, aliasing can not only produce high frequency mirror tones, but also muddy out the lower end, even reflecting back down under the fundamental and into DC land. I don't understand this artifact very well but you can see it on a spectrograph fairly clearly.

[xoxos]

Being somewhat of a beginner in the practical know-how department of sound design, mixing etc., I know how aliasing works in theory: i.e., going from continuous to discrete in the time or frequency domain will yield periodisation in the other.

drop a spectrum analyser on your instrument, raise the pitch. see any peaks moving downwards, that's aliasing.

you can hear aliasing if you know what you're supposed to be hearing without it (eg. elementary waveforms).

two considerations:

1) aliasing is often desirable (eg. downsampling and bit ops, insertpizhere)

2) aliasing is less readily tangible to the inexperienced. it's a de rigeur way to detract from the perceived desirability of products you do not wish to be prevalent in the marketplace, because many of the readers are looking for some experienced poster who can discern what they can't yet.

[foosnark]

In all cases, aliasing sounds bad.

In typical cases, aliasing sounds bad.

Some of us adore our downsampling and bitcrushing and ring modulation and other shenanigans.

[fluffy_little_something]

So, what can we do to minimize it? Say, a synth has aliasing, can I minimize that by changing certain settings?

(PS: A woman on this forum, miracles do happen Glad you are here, Maria.)

[Music Engineer]

Say, a synth has aliasing, can I minimize that by changing certain settings?

if such settings are available: interpolation mode, oversampling factor

the latter can be done even if the synth doesn't support it, by using a higher project sample rate in the host and/or an oversampling subhost.

in samplers: use multisamples to avoid high transposition factors

in synths that support import of single-cycle waveforms, you could use waveforms that are more bandlimited (but this may create a dull sound)

in FM synths, it might be possible to reduce the modulation index for higher keys with a (negative) keytracking parameter.

use less distortion.

...just some ideas. it really depends on the synth's architecture.

[earlevel]

There a fairly quick aliased sawtooth sweep here, but I have a much better example in a ~2 minute video on wavetable oscillators that I'll post in an article very soon.

As others have said, aliasing is most obvious in things like pitch bends. For instance, with naive (non-oversampled) digital guitar distortion, it may be hard to hear aliasing in all that distortion grunge...till you bend a note upwards and hear tones going downwards.

For non-clipped synth sounds and traditional synth waveforms, where amplitude drops for higher harmonics, aliasing is much more obvious on higher notes. While this may be obvious enough, since they are closer to the dreaded Nyquist frequency, it's helpful to think about the time domain view as well. Higher notes have shorter cycles—fewer samples per cycle. If sharp (as in jagged) feature in the waveform is misplaced by a half of a sample in the bottom octave it's off by 1 in ~4000. But if it's off by that much for a note four octave above middle C, the same error is off by 1 in ~5. There's just less resolution per cycle up there.

[BertKoor]

My question is: How do you recognise this by ear, given that aliasing is dependant on the character of the original signal and the sampling frequency?

The simplest "litmus test" for aliasing is to play a note and move the pitch wheel. If you hear harmonics going in the opposite direction, that's aliasing right in your face.

(whoops, this advice was given before. That happens if you reply before reading any further )

[GeckoYamori]

I actually usee aliasing for sound design purposes in FM synthesis. At least, I'm pretty sure it's aliasing. Usually I modulate at extremly high frequencies to get certain metallic, cymbally timbres and these can sound fundamentally different per each semitone which I suspect is due to aliasing artifacts mucking things up in the modulation, and sounds quite interesting. I tried playing these sounds at higher sample rates and the effect became more subdued.

[Ingonator]

In all cases, aliasing sounds bad.

Not in ALL cases. In older synths like the PPG Wave 2 it's part of the sound.
Waldorf is even using an emulation (!!!) of the aliasing in the different PPG Wave models (2.2 + 2.3) in the PPG Wave 3.V.

I agree that it is bad in an analog emulation as such things should not happen in an analog synth.


Ingo

[Gamma-UT]

In all cases, aliasing sounds bad. Our ears are especially sensitive to it because the aliased frequencies are not harmonically related to the source tones, making them stand out. It's like intermodulation distortion in that respect - even a little bit is offensive to the ear.

Tell that to owners of the SP1200.

[IrionDaRonin]

In all cases, aliasing sounds bad.

I have to be the one of the few that like aliasing sound, or at least i can support it.
To put an example, Toxic Biohazard, i like the "lofi" feeling.
It sounds alike to Tron Legacy soundtrack.

If a synth got aliasing, and it works for your actual song, it should be ok to use it.
All depends always on what we are looking for our tracks/style/album.

Have a nice day