Why do music modules sound different on anything other than Amiga?

[mc6809e]

Originally Posted by FromWithin

It's not a belief. It's a mathematical certainty.

Uhg. So you think the sinc function is simple? It's not even finite!

But of course we don't really need the entire sinc function. But we still need enough of it to convolve it with the sample. And how much of sample? The entire sample!

Once we've done that, then we can say with certainty that we've perfectly reconstructed the original, band limited signal.

Don't think that because perfect reconstruction is possible that you're guaranteed to get it with some generic low pass filter. That's not what what the reconstruction theorem says.

[roondar]

If we're going to talk mathematical certainties and audio, there are at least three things we really have to keep in mind. And I feel these things are glossed over far too easily in any discussion over audio.

The first is that human hearing is actually severely limited: There are plenty of frequencies we simply cannot hear. Human hearing has a very clear and proven upper and lower limit for which frequencies we can actually perceive. There is also, despite plenty of controversy and a whole bunch of studies into it, as of yet still no conclusive scientific evidence to support the notion that we may only hear up to a certain frequency but still somehow notice (f.ex. due to harmonics) it if these higher frequencies are missing or distorted. Note here that we can perceive lower frequencies than we can hear through other means though - primarily because such low frequencies can actually move parts of our bodies slightly and that can be perceived (given enough volume).

The above is part of the reason why compressed audio is really rather hard to recognize vs uncompressed audio (for a fun test how well you can manage, check out this blind test: https://www.npr.org/sections/thereco...=1550140442261).


The second is that all audio playback devices ever made are completely incapable of actually playing back audio at anything even resembling 'perfect' quality. There are no amplifiers which do not distort. There are no speakers with infinite frequency range, or any that have fully flat frequency responses. There are no audio playback devices that play back audio 'perfectly' (this includes digital devices, which invariably introduce jitter). Nor are there audio formats that even come close to storing audio 'perfectly' to begin with. And that is not even counting the losses from recording and the imperfections of microphones.


The third is that it has been (mathematically) proven that, given the parameters required for output, you can in fact sample audio in such a way that you can't tell the difference between the sampled version and the original - within these parameters at any rate. This is the essence of the Nyquist–Shannon sampling theorem - which states you need twice the sampling rate as the highest frequency you wish to store. As it happens, given the limits of the human ear a ~40KHz sampling rate is enough to not lose any frequency you can actually hear (given that only very, very few people can hear anything over 20KHz).


The upshot here is that you may well describe a 'perfect' wave form using a mathematical formula, but you'll never ever get such a wave form stored nor played back using any equipment that can be build. The far more interesting question is how you can make something that gets close enough as to make it impossible for us to actually hear it isn't 'perfect' (as in does not appear to lose anything over what the microphones used for recording output).

And while the Amiga definitely won't reach that standard using Paula, all signs point to something as simple as a competently implemented digital system with somewhere between 16 and 24 bits of resolution (as in more than 16 but less than 24) running at around a 44-48KHz sample rate should in fact be indistinguishable from what the original microphones picked up. Even for people with really good hearing.

Such a system pretty much leaves all of us Amiga fans in the dust as most of us are over 35 and male - which means our hearing likely tops out at 14-15KHz (i.e. 30KHz samples ought to be enough for us oldies)

[Nishicorn]

Real Amiga and emulation can sound surprisingly different. I didn't think it would have a big difference, but when I used an Amiga in one of my music projects, I had to actually (sadly!) switch to using emulator instead, as the real Amiga gave WAY too much bass and threw everything off!

The reason was, I thought I could just compose on emulator, and then use the real Amiga for the final audio capture, but I was so wrong.. there's something about 'the real thing', especially when it comes to these quirky, old computers, sound chips and systems, that really makes a difference and brings a bit more life compared to cold emulation.

Emulation is a fantastic thing, it may actually preserve many machines way beyond their physical years (after all, everything physical will change forms at some point, crumbling to dust, for example), maybe even thousands of years into the future (who knows..)

But there's just -something- special about using a real Amiga, or even a real OPL chip that just brings a little bit more excitement into the mix. And it will always be a different experience, even when logic would dictate it should be 'similar enough'.

[Toni Wilen]

If sound is different (emulation vs hardware), it has to be due to op amp stage in Amigas. Paula is pure digital, I don't think it can change sound that way.

Someone who knows this stuff should check the op amp circuitry if it does something more than simply amplifying (and filtering) Paula outputs. (Note that op amp circuitry also is not identical in all models)

[Thomas Richter]

Originally Posted by Toni Wilen

If sound is different (emulation vs hardware), it has to be due to op amp stage in Amigas. Paula is pure digital, I don't think it can change sound that way.

Indeed. Paula's A/D is probably quite simple, and it may vary a bit due to lack of proper calibration, though its frequency characteristics is most likely quite linear.

The Op-AMP stage includes essentially four filters: An initial low-pass that cannot be turned off, then the control-able 8K cut-off filter which you all know can be disabled by the same line that controls the power LED, and another C-RC combination at the output that first removes any DC bias, and another low-pass filter that cannot be turned off directly at the output.

The cut-off filter is controlled by a FET which is itself not a linear element, so some additional distortion comes in at this stage.

Anyhow, it is clearly not a flat frequency response we have here in the output filtering stage of Paula.

[Toni Wilen]

Also component values may have drifted in op amp section, power supply may also affect sound. I assume test was done using same amplifier with volume level matched. Without it test would be invalid.

I'll try (if I am not lazy enough) to do some tests. Play different frequencies with same amplitude and check output with a scope. And other similar tests.

[commodorejohn]

Originally Posted by Toni Wilen

If sound is different (emulation vs hardware), it has to be due to op amp stage in Amigas. Paula is pure digital, I don't think it can change sound that way.

Except that Paula's DACs have long been noted for their odd, non-standard response curve (and there was a detailed discussion as to the nitty-gritty of that upthread,) so what you get out of it when you feed a sequence of numbers into it is not the same thing as what you'll get feeding that same sequence into the DAC on a modern PC sound card.

That said, it wouldn't surprise me to learn that the op-amp stage is also quirky and not studio-grade precision engineering, and the filters which were also discussed upthread definitely play a large part in shaping the characteristic sound as well. But just because something is "pure digital" on the backend doesn't mean it can't undergo strange mutations on its way to transforming into an analog output signal.

[Toni Wilen]

Yeah but I don't think non-linear dac can cause reported differences in audio.

[commodorejohn]

It absolutely can. Getting a louder signal level where you intended a softer one or vice-versa changes the dynamic range of the signal and distorts the resulting waveform relative to the intended result. That can make a huge difference in the sound, especially when you're only starting with an 8-bit sample's worth of dynamic range in the first place. Compressing and distorting the waveshape can easily give the kind of punchy, crunchy character that the Amiga is famous for.

[roondar]

I've personally never noted that much of a difference between WinUAE and real Amiga audio output. With that in mind, I found a YouTube channel a while back that records music of real Amiga's and puts them through a Dolby headphone style treatment.

One of the videos he did compared real A500 & A1200 output (with and without filter) with WinUAE. You can listen/watch here: [ Show youtube player ]

I can't really hear the difference myself but... It is processed by his Dolby thing so that may distort things. Maybe it's useful in some way regardless.

[commodorejohn]

I wouldn't rely on YouTube for audio comparisons anyway; looks like there was a lossless recording up at one point, but it gives me a 403 error now.

[Dot77]

Originally Posted by roondar

...

One of the videos he did compared real A500 & A1200 output (with and without filter) with WinUAE. You can listen/watch here: [ Show youtube player ]

I can't really hear the difference myself but... It is processed by his Dolby thing so that may distort things. Maybe it's useful in some way regardless.

Lossless flacs here https://bit.ly/2TTBcKR (zipped, 74MB)
Includes pure unprocessed (apart from gain matching), stereo joined and Dolbyfied versions. Link is active until the end of March.

[roondar]

Originally Posted by Dot77

Lossless flacs here https://bit.ly/2TTBcKR (zipped, 74MB)
Includes pure unprocessed (apart from gain matching), stereo joined and Dolbyfied versions. Link is active until the end of March.

Thanks!

I've downloaded them and listened to them for a bit.

To be honest, I still can't hear the difference. Obviously that doesn't have to mean no one can, but I can't regardless. Which means that at least for me, WinUAE is good enough.

[daxb]

I checked the lossless flac files and can't hear a difference between UAE and real. But I would say that the source chip tune is a very bad choice. Some clean single samples with different frequencies would be better to compare or simple test tones.

[Nishicorn]

Try with something that has a lot of 'sub-bass' - it can sound 'perfect' with WinUAE, and then you play it with real Amiga, and there's BOOMING BASS that overpoweres everything.

Could be just my Amiga or sound system, though .. but it seems real Amiga 1200 outputs a bit more bass than WinUAE, at least for that particular music project.

It was actually a kickdrum I made with a real 8580 (which is famous for its 'sub-bass' quality when the filter cut-off is ever-so-slightly above zero), that I converted to a sample format Protracker can understand, that messed everything up if I used the real Amiga, so I had to use WinUAE instead.

[Hewitson]

I also find a real Amiga sounds considerably different (better) than an emulator (or UADE) running on a PC.

Particularly in the bass but differences can be noticed in other areas.

[DrFruit]

Sounds good to me in hippo and protracker 2.3a, I’m running them on a real a500 and winuae through a naim audio setup and they sound the same?


Sent from my iPhone using Tapatalk