Snes Sound

>>3839672
A few reasons:
1) Without going into to much detail, the decoded drops 1 out of every 4 bits.
2) Unconfigurable gaussian filter applied on output
3) APU coprocessor memory space made fitting code, sound, and music in the same small 64k memory space challenging. You could get around this with modest effort and careful sample compression, but not all companies had that kind of talent or time.

>>3839675
First post best post.

Now let us commence 60 replies about how the Genesis sounds like robot farts.

>>3839683
remember when your robot farted and it sounded like kid chameleon? good times

>>3839683
BRAAAAAAAAAAPPPP

>>3839672
Same reason every developer reused that same ugly MODE 7 gimmick in all snes games.

>>3839672
I've heard that all licensed SNES games were forced to use the same sound driver, which soaked everything in reverb and occupied most of the 32 audio channels the SPC 700 was supposedly capable of.

>>3839731
>forced to use the same sound driver,
That's retarded. Don't repeat stuff you have no idea about. While some games share sound drivers, there is enormous diversity in what you find in the wild. For example, Tales of Phantasia had a sound driver that could stream voice samples to the APU and cross-fade two songs, something that they were proud enough of to advertise when you start up their games.

>>3839683
BRRAWWWWWWWWWWW

>>3839675
I would like to know more about reason A, if you would have time to get into those details

>>3839769
SNES audio uses a sound format called Bit-Rate-Reduction (BRR) to encode a pulse-amplitude-moduluted (PAM) wave.

Essentially a sound wave's amplitude is sampled at regular time intervals to produce a series of samples that represents the sound as a PAM wave. This is about the most direct way to encode sound there is.

BRR seeks to compress the sound, because memory was expensive back then. The way BRR does this is by splitting a sound into 16-sample chunks. Each chunk has a 1-byte header, followed by 16x 4-bit signed integers (ranging from [-8,7]) representing the samples. The header dictates the level of the sound base 2 (among other things).

amplitude = sample * 2^level

The idea is that the amplitude of a sound wave should be roughly the same magnitude over so few samples, so
by chunking similar data, some of that data can be shared and thus compressed, the level being the shared data in this case.

So what about my statement about dropping 1 out of every 4 bits? By some retarded undocumented design decision, the least significant bit is dropped by the samples upon decode, thus they only range from [-4,3], halving the fidelity.

It's slightly complicated by the header also specifying filters in which the current sample is calculated from previous samples. But seriously, wtf Ken Kutaragi? All that wasted memory and more muffled sound to boot. Fortunately emulators can renable the 4th bit.

>>3839781
Your description of BRR makes it sound identical to ADPCM.

>the least significant bit is dropped by the samples upon decode, thus they only range from [-4,3], halving the fidelity.
Probably to save hardware cost in silicon, much like earlier Playstation models always read textures at 15bpp (even if the source is 24bpp) to save hardware cost, causing lots of banding.

>>3839781

That sounds like a LOT of overhead

>>3839683
If anything wouldnt the reverb from the snes sound more like farts?

>>3839793
It's essentially a proprietary implementation of ADPCM. I actually meant to put PCM instead of PAM (I've never called it that before, but was blanking on PCM in the wee hours of this morning)

>>3839731
>32 audio channels
Lol, even the PS1 has only 24 channels.

>>3839675
You basically had the choice between a lot of unclear samples or a few clear samples.
https://www.youtube.com/watch?v=wRHyH824KIM

>>3839683
ooooh, that was a wet one!

https://www.youtube.com/watch?v=g2oyweqZ7WM

>>3839683
Thank you.

>>3839683
Sega sounds okay, you just had bad experience with shitty games.

https://www.youtube.com/watch?v=ukjuhIG_rf8

>>3839801
I'm not out-right rejecting your theory, but only so you can dig up some examples.

>>3839969
I want Doom 32x to fart on my face

>>3839781
>So what about my statement about dropping 1 out of every 4 bits? By some retarded undocumented design decision, the least significant bit is dropped by the samples upon decode, thus they only range from [-4,3], halving the fidelity.
>It's slightly complicated by the header also specifying filters in which the current sample is calculated from previous samples. But seriously, wtf Ken Kutaragi? All that wasted memory and more muffled sound to boot. Fortunately emulators can renable the 4th bit.

Are there any comparison videos/mp3s of this?

Can you describe how much improvement there is or how the sound changes?

Which SNES emulators have this feature?

>>3839672
That's just how low-resolution samplers sound.

>>3839675
All of the samples were compressed. SNES used Adaptive PCM as opposed to standard PCM. The Genesis could play standard PCM, but 16-bit stereo samples would have been out of the question due to file size.

File: 1431477579517.jpg (171KB, 1520x1520px) Image search: [Google]

171KB, 1520x1520px

>>3839675
>drops 1 out of every 4 bits
How is that even acceptable at all? Like what the fuck.

>>3839672
What's the Sonic Spinball menu theme of SNES? I.e. the tune that's all the worst of SNES in one package?

>>3840967
I think he meant encoded, and this is how lossy compression works.

Because you're using a PAL system.

>>3839672
Still better than PC: https://www.youtube.com/watch?v=L1pCxXrorKM

>>3841057
But that uses a sound chip from 1985 or so.

>>3841017
No, I'm saying that the waveform encoding literally has dead bits in it that are never used. For every 9 byte BRR sample, literally 2 bytes go unused because the hardware never uses them. This is not noted in the official dev manual.

This is emphatically not a lossy vs lossless compression issue.

>>3841057
https://www.youtube.com/watch?v=68clKOeaWp8

>>3839781
Thanks for the detailed answer.
Still there's some stuff I don't get. Wouldn't the lost bit be adjacent to a bit that has the same value thus being easy to recreate?(as long as you keep track of the position it is, maybe that could also come in the header and be shared across the 4 4 bits chunk) why is it always the least significant bit?

>>3839672
Better sound than genesis