The aliasing thread

[bmanic]

sorry if I sound jaded, but it really seems like a non-issue. certainly not something that should matter to the average person.

Why do people find it so difficult to understand that in isolation, a small quality difference is easy to dismiss (aka, "there's such a small difference in quality, nobody will notice") but once you use 50 of those samples it will ADD UP. And then people ask " why does this sound so home made "..

The whole point with professionalism and making something sound (taste, look, feel etc. it's the same in most other professions) really good you HAVE TO pay attention to the details as the details add up!

Key sentence: It adds up!

Another note: Don't underestimate the audience. Depending on the genre there are some extremely ear trained audiences out there. I'm sure the average teen that listens to brittney wouldn't give a f**k about audio quality as long as the drums are punchy enough and she wears minimal clothing but there are tons of adults that listen to stuff that depends quite a bit on production.

Cheers!
bManic

[Fire Sledge - Ohm Force]

Code: Select all

// fsse_interpolate_sinc ()

Fantastic, René !

As I understand the code, you're using some kind of polyphase impulse (SPASS*12 taps), linearly interpolated, then convoluted with the signal. Are you using any extra-oversampling, or mip-mapping ?

-- Laurent

[René]

As I understand the code, you're using some kind of polyphase impulse (SPASS*12 taps), linearly interpolated, then convoluted with the signal. Are you using any extra-oversampling, or mip-mapping ?

Exactly. Oversampling is used in the impulse calculation (about 256x iirc), which is precalculated. There's no oversampling in the convolution process.

I just keep two arrays, one with the impulse itself, and another with the differences for subsequent points which speeds up the impulse interpolation process.

The impulse tables are calculated as standard windowed-sinc, but in lower quality modes (7 and 15 points) I've found that using a Remez exchange generated impulse results in better distribution of the noise across the frequency range. So the impulse doesn't exactly resembles a sinc, but it has some 'extra nodes' on it.

My design goal was:

1- having interpolation noise < -96dB (16-bit limit)
2- having aliased images noise < -96dB (the damn arguru made me push it to -144dB now)
3- having uniform (or almost) noise vs. frequency distribution.

'Point-connecting' interpolation methods (bicubic, lagrange, hermite, etc.) have the unwanted side effect of having exponentially more interpolation noise when frequency increases. Sinc has almost the same noise figures across all the spectra (besides the bandlimiting effect).

-René

[tony tony chopper]

And then people ask " why does this sound so home made "..

The whole point with professionalism and making something sound (taste, look, feel etc. it's the same in most other professions) really good you HAVE TO pay attention to the details as the details add up!

mmh, I don't neglect antialiasing, it's important..

BUT you'll also notice that ALL of the said professional samplers are at the bottom of the chart.
When people ask why it sounds home-made, it's certainly not because of aliasing. It's because it's not well mastered, that is, equalized and compressed properly. Now with threads such as these, people will think that a perfect antialiasing will give them 'that sound'. Do you think it'll be the case?

[René]

#undef professional

I'm not exactly sure what aspects of a music track make people call it 'home-made'.

There for sure might be a hundred things which could be characterized with that expression. Certainly, interpolation noise might be one of those.

Now, here's a render of Gigastudio, the 1995 sampler.

http://www.rgcaudio.com/images/gs.gif

Note that it's not bandlimited, but still has a pretty low interpolation noise. It would be a very strong contender. And it's 10 years old.

I don't know what sampler most 'professionals' use. However, I do know a gazillion professionals using Gigastudio. In the field I'm most familiar with (natural, orchestral instruments) there's just nothing on earth which can compete with it. Too bad it's not plugin-able.

Some nice libs for <other samplers here> have appeared lately. But every good-sounding lib I can think of uses chromatic sampling, so there's no resampling.

-René

[george]

People will think that a perfect antialiasing will give them 'that sound'.

Apparently FLS cares about "that sound" since users are able to render offline using sinc256, so don't you think it's pretty well covered already?

Now, here's a render of Gigastudio, the 1995 sampler.

That's a superb result considering how many years has been out. Can I put that render at the sample page?

[René]

That's a superb result considering how many years has been out. Can I put that render at the sample page?

Of course, is that gif ok or you prefer me to send you the rendered wav?

-René

[george]

Of course, is that gif ok or you prefer me to send you the rendered wav?

I prefer to get the wav, thanks.

[tony tony chopper]

Apparently FLS cares about "that sound" since users are able to render offline using sinc256, so don't you think it's pretty well covered already?

I care about aliasing and I care about 'that sound', but I don't mix both. I'm well aware that antialiasing brings a 'clean sound', but it's not 'that sound'. For 'that sound', we have mastering plugins. Fortunately or unfortunately, most people want their song to sound like popular radio or CD mixes, and the reason their songs don't is certainly not aliasing.

[Urs]

Just popping in (time to go downstairs and have a beer ).

I agree with gol. Look at the Instruments forum. They typically measure a synth's quality by the Virus. In addition to some effing non linearities, that also means quite some aliasing

René, it's pity I can't read Intel SIMD asm...

Cheers,

Urs

[george]

Look at the Instruments forum. They typically measure a synth's quality by the Virus. In addition to some effing non linearities, that also means quite some aliasing

Isn't instruments forum the place where "tarnce", "sounds like a virus" and "the best 303" topics pop more often? It has always been very synth focused.

FYI, site has been updated with GigaStudio 2.5 graphics and audio (thanks to René).

[WilliamK]

Guys, one thing that I don't understand. The first graphics shows a "hole" after 20.000hz. While all others shows a lower value for those. Doesn't that means that Sinc-Interpolation is limiting the HZ ceiling while others just mirror the information? Is that a good or bad thing? Hope that I explained it right.

Wk

[bmanic]

Duh, gol,

of course you need to EQ etc. properly to get a better sound but how the hell are people supposed to do those more complex things if they can't even hear the bad effects of aliasing? The whole point is to learn how things sound and WHY they can sound certain ways. Also, depth, character, punch and warmth (yes, there IS such a term in describing music/sound) does not come simply from EQing. Do you really think that a crap violin can be made to sound nice with simply using EQ? No, the GIGO (garbage in - garbage out) rule applies very much to music and this is why aliasing and other unwanted artifacts are bad. Once you have it in your music there's no way EQ or anything else can help you get rid of it. Hence it's worth pointing out and getting rid of.

Like I said, professionalism is about the details but of course untill you can even detect details and know what to look for you need to learn the tools and practise the ear-brain connection. There are plenty of people who can produce tracks with even frequencies and make it sound ok but those tracks can still sound dead and cold caused by using inferior tools to produce the sounds to begin with.

- bManic

EDIT: just to specify. I'm now talking about aliasing in samples. NOT synths.

[bmanic]

Apparently FLS cares about "that sound" since users are able to render offline using sinc256, so don't you think it's pretty well covered already?

I care about aliasing and I care about 'that sound', but I don't mix both. I'm well aware that antialiasing brings a 'clean sound', but it's not 'that sound'. For 'that sound', we have mastering plugins. Fortunately or unfortunately, most people want their song to sound like popular radio or CD mixes, and the reason their songs don't is certainly not aliasing.

So, lets say I make a song using some extremely bad quality accoustic instruments. Putting some mastering plugins suddenly can make my song be radio quality? I don't think so. Again, it's in the details of choosing QUALITY instruments to begin with. This way you have much better chance of getting a professional sound.

You can't make a good soup with rotten meat, no matter how much spices you add.

- bManic

[Blue Days]

autloc: About the amiga, it used variable-rate DAC sampleplayback, so the antialiasing filter actually filtered out aliasing above samplerate. Not wrapped aliasing due to resampling. Also it's much easier to deal with since it doesn't require interpolation either.