Hertz Overload - new demo for AGA Amigas

[saimo]

Originally Posted by pandy71

Of course i can prepare samples without such boost and also sample length to 10 seconds to address your setup specification.

Yes, please: just unfiltered, noise-shaped samples.

[Karlos]

Originally Posted by Gorf

Yes - I linked to it already here in the thread.

I must've missed it.

[pandy71]

Originally Posted by saimo

Yes, please: just unfiltered, noise-shaped samples.

Hi Saimo, exactly as you requested: https://eab.abime.net/zone/For_Saimo_2.7z

[saimo]

Originally Posted by saimo

I'll have to redo the tests. Now I just did quick and different one that seems to indicate that things are worse than I reported in my previous post, but I'm too exhausted and I might just be messing things up once again. Tomorrow or in the next days I'll return to it.

As promised, I made some more tests. I went for a (hopefully) more reliable method, though: I modified the 8-bit player (sAQA0) to increase the period in steps of 1 from 37 to 64 every 256 kB played and let it play the test tone with volumes 1, 19, 32, 53 and 64.
I didn't consider smaller periods as I can sample at most at 192 kHz (and period 37 gives a frequency of about 96 kHz, which is already overkill). Also, I used a 1000 Hz sinus tone sampled at 55420 Hz, as it looks like the 13855 Hz square tone was hard for Paula to digest (but I didn't really investigate further).

Well, apart from a few hiccups (especially at the smaller periods), it turns out that Paula does a pretty good job (although not perfect) at regulating the volume (as usual, click the pictures to see them in full size):





samples and pictures download

Finally, I have added volume and filter command line options to the sAQAx players.

P.S. I have removed the files I had previously uploaded as they were cluttering the server and were obsolete anyway.

EDIT: I also fed the modified player the Sony music snippet and it worked very well; I didn't include it here and in the archive as at first I thought that it was more a curiosity than else; but now I regret it

[saimo]

Originally Posted by pandy71

Hi Saimo, exactly as you requested: https://eab.abime.net/zone/For_Saimo_2.7z

Great, thank you! Tomorrow I'll give them a spin.

[saimo]

@pandy

Originally Posted by pandy71

Hi Saimo, exactly as you requested: https://eab.abime.net/zone/For_Saimo_2.7z

Tests made.
For practical reasons (i.e. floppies swapping - I have a Gotek, but still...), I have used only the last 7.5 seconds.
For comparison, I also exported the same chunk with Audacity.

Results:

Audacity = made with Audacity
8/14/16  = 8/14/16-bit
ND       = No Dithering
SD       = Shaped Dithering
sampled  = Amiga 192 kHz sampled output

8-bit:
BeeMoved_2requant_ssstrong64k-sampled: -1.910 dB, clearly audible fine noise
BeeMoved_improved-e-weighted-sampled:  -2.235 dB, audible, but weakest noise
Audacity-8-ND-sampled:                 -0.942 dB, strong rough noise
Audacity-8-SD-sampled:                 -0.928 dB, strong rough noise, comparable to Audacity-8-ND-sampled

14-bit
BeeMoved_2requant-sampled:             -2.364 dB, barely audible noise
Audacity-14-ND-sampled:                -0.913 dB, barely audible noise
Audacity-14-SD-sampled:                -0.935 dB, a bit more audible noise

The provided samples, noise-wise, give better results than Audacity, but suffer from a sizeable dynamic range loss. It turns out that the best-sounding result is the non-dithered non-noise-shaped 16-bit sample produced by Audacity.

samples download

I'm more and more convinced that, in general, the best solution is either 14-bit without noise shaping, using 55420 Hz sampled rate instead of 64489 Hz (thanks to the absence of noise shaping), or AM 8-bit samples at 55420 Hz again.
For the demo I'll stick to the crazy 14-bit 70937.9 Hz, as yesterday's tests showed that Paula handles graciously a volume of 1 at that frequency, but I really think that AM should be explored...

@Karlos
If I understand correctly, you happen to have (since a long time) a ready-made converter that produces AM data - is that so? Is it public or, if not, could you simply convert a file so that I can check the resulting on the A1200, please?

[Karlos]

@saimo

Just this: https://github.com/0xABADCAFE/paula-hdr which just contains simulation of the mechanism (it's also intended to compile on generic linux)

However, @paraj posted an implementation in the thread earlier that does it all, runs on an amiga and includes a reference paula output implementation.

[saimo]

Originally Posted by Karlos

@saimoJust this: https://github.com/0xABADCAFE/paula-hdr which just contains simulation of the mechanism (it's also intended to compile on generic linux)However, @paraj posted an implementation in the thread earlier that does it all, runs on an amiga and includes a reference paula output implementation.

Cool, I'll have a look. Thanks (and also to paraj, of course).

[Karlos]

Originally Posted by saimo

Cool, I'll have a look. Thanks (and also to paraj, of course).

I'm planning to use this same technique for a replacement TKG sound engine. Fun times ahead.

[saimo]

Originally Posted by paraj

Found old quick&dirty test source for that. Assumes 16-bit LE samples.

Thanks for this. I had totally overlooked it.
I just tried it (after compiling it with VBCC) on WinUAE. For some reason the playback is slow (at first I had erroneously used a stereo sample, but then a mono one didn't fix the issue).
Anyway, no worries and no action required - I just dropped these lines for acknowledgement and feedback. I was just hoping to be able to make a quick and meaningful test with AM, but for my player I need a suitable converter, so I'll write one sometime tomorrow or in the next few days.

[saimo]

Originally Posted by Karlos

I'm planning to use this same technique for a replacement TKG sound engine. Fun times ahead.

Guess what? Only now I noticed (consciously, at least) your avatar. That game's baddies and bugs and code and algorithms and mods etc. must be visiting you in your sleep quite often

[paraj]

Originally Posted by saimo

Thanks for this. I had totally overlooked it.I just tried it (after compiling it with VBCC) on WinUAE. For some reason the playback is slow (at first I had erroneously used a stereo sample, but then a mono one didn't fix the issue).Anyway, no worries and no action required - I just dropped these lines for acknowledgement and feedback. I was just hoping to be able to make a quick and meaningful test with AM, but for my player I need a suitable converter, so I'll write one sometime tomorrow or in the next few days.

That's strange (assuming you used specified the correct period ), since I tested it with WinUAE and real hardware alike. I put exe+test sample in the zone (period = 124).

[saimo]

Originally Posted by paraj

That's strange (assuming you used specified the correct period ), since I tested it with WinUAE and real hardware alike. I put exe+test sample in the zone (period = 124).

Just tried it: it works with the provided sample and so does the executable compiled from the source. Also, both executables work with a period of 55 with the provided sample. However, they also both fail the same way with this sample here: https://www.retream.com/_temporary/m.wav https://www.retream.com/_temporary/m.raw (mono 16 bit signed LE). It doesn't seem to be a period issue, as the playback duration is correct (7.5 s). Don't know... but, really, nothing to worry about

[paraj]

Originally Posted by saimo

Just tried it: it works with the provided sample and so does the executable compiled from the source. Also, both executable work with a period of 55 with the provided sample. However, they also both fail the same way with this sample here: https://www.retream.com/_temporary/m.wav https://www.retream.com/_temporary/m.raw (mono 16 bit signed LE). It doesn't seem to be a period issue, as the playback duration is correct (7.5 s). Don't know... but, really, nothing to worry about

A period of 55 is too low when using DMA like my code does

[saimo]

Originally Posted by paraj

A period of 55 is too low when using DMA like my code does

Ah, alright. I'm so used now to making tests with stupidly small periods that I just typed commands away in the shell without thinking
(Strangely though, with the sample you included the playback does get sped up with 55 - but I didn't time the duration).

[Karlos]

The little endian stuff is just an artefact from prototyping the original code on x64, there's no algorithmic necessity for it.

[saimo]

Originally Posted by Karlos

The little endian stuff is just an artefact from prototyping the original code on x64, there's no algorithmic necessity for it.

I see. Well, that's no problem (even my converters use LE as they take WAV files as input).

[saimo]

Having improved the converter and having improved the music mix, I'm about to release an updated version of the demo.
Before doing that, though, I made one last test to double check how Paula copes with volume 1 at period 50 (the one used by the demo). What my limited means report is that Paula does indeed a very good job (click for full size):



test tone download

Also, the outcome is quite fine to the ear, so the updated demo is going to be out soon.

[saimo]

... and now it's released. As anticipated, the music is better now - which is quite relevant for a music-centric demo

CHANGELOG

v1.2 (10.6.2024)
* Improved streamed music: made arpeggio and strumming guitars stronger in the mix; managed to eliminate some more peaks, thus exploiting more the resolution and increasing the overall quality; made an overlapping solo guitar enter from the middle-left and then move to the center gradually; converted data with an updated and more precise custom converter.
* Increased the volume level of the arpeggio guitar in the tracked music.
* Improved a few pixels.
* Extended and fixed manual.

https://www.retream.com/Hertz_Overload / https://retream.itch.io/hertz-overload

[ Show youtube player ]

[Karlos]

That sounds great, but I can only hear it via YT, so I'm not getting the true experience.