The originals sound better. The aliasing provides a crunchiness and sharpness to the final output that drives emotional energy. That zero mission rhythm isn't intended to sound smooth and soft, the driving hard beats are an emotional tool for eliciting anxiety and anticipation from the player.
But this is a bit like those who use smoothing filters. It's ultimately about taste, but it should be recognized that unless the filter is attempting to accurately recreate the original hardware of the era then the original design intent is not being adhered to, and so something may be lost in the "enhancement".
I don't think so, I think you're just getting a high end that isn't in the original audio. In the places where there are high frequencies the aliasing and the hiss just gets in the way.
> The originals sound better. The aliasing provides a crunchiness and sharpness to the final output that drives emotional energy.
In the mid-1980s the first really affordable sampler was the Ensoniq Mirage, which used the Bob Yannes-designed ES5503 DOC (Digital Oscillator Chip) to generate its waveforms. It played back 8-bit samples and used a fairly simple phase accumulator that didn't do any form of interpolation (I don't count "leftmost neighbour" as interpolation). Particularly when you pitch it down, you get a rough, clanky, gritty "whine" to samples, that the analogue filters didn't necessarily do a lot to remove.
Later on they released the EPS which had 13-bit sampling. Why 13-bit? I don't know, I guess because the Emulator I and II used 8-bit samples but μ-law coding, giving effectively 13-bit equivalent resolution. It also used linear interpolation to smooth the "jumps" between samples, and even if you loaded in and converted a Mirage disk the "graininess" when you pitched things down was gone.
I'm currently writing some code to play back Mirage samples from disk images, and I've actually added a linear interpolator to it. Some things sound better with it, some things sound worse. I think I'll make it a front panel control, so you can turn it on and off as you want.
It's really close for me. I listened to the accurate version, then the enhanced one, and my first thought was "oh, yeah, this sounds better."
Then I listened to the accurate version again, and thought "wait, never mind, this one sounds better."
After going back and forth a few times, I think I still agree the original/accurate one is better, but it's pretty close. I really encourage people to listen for themselves.
For what it's worth, I have little to no personal nostalgia for the Game Boy Advance.
This is fair, I should have made it clearer that this is all subjective. For what it's worth I have all of this disabled by default in my own emulator because I think default settings should always err towards accuracy when that's a question.
I personally do prefer the interpolated versions in most cases because to me the extra high-frequency information just sounds like noise that makes it harder for my brain to process the underlying music. But clearly many feel differently!
Well, we can't just say 'original = intent'. The original artists presumably did the best job of expressing their intent as far as possible in the medium at the time, but that doesn't mean that this necessarily is the best expression of their intent ever.
It's like saying you can only watch the Simpsons with the exact late 1980s / early 1990s ads that they originally aired with, and everything else is sacrilege.
But without asking them it's pure conjecture. I don't think trying to retcon the best expression of their intent needs to be used to justify this project, either. Sometimes it's fun to see if you can build an improvement on what exists, even if it's a vehicle for learning about DSP or whatever domain the learner is in.
I think it's not so much that one sounds better or the other.
The "uninterpolated" one is incorrect.
The "interpolated" one is incorrect.
The uninterpolated one has sharp square edges, which isn't correct. The GBA has a 12dB/octave filter at around 12kHz (IIRC) on the output, which the uninterpolated simulated output doesn't appear to have. This would knock the corners off a bit and make it "smoother" and less hissy, but would still have quite crunchy low frequency sounds.
The interpolated one smooths things off excessively, and while it doesn't really have much less spectral energy high up, what's there is in the wrong place.
This is great stuff… basically, an easy way to get much higher quality audio out of a GBA emulator.
I’ll add some context here—why don’t more games run their audio at 32768 Hz, if that’s such a natural rate to run audio? The answer lies in how you fill the buffers. In any modern, sensible audio system, you can check how much space is available in the audio buffer and simply fill it. The GBA lacks a mechanism to query this. Instead, what you do is calculate this yourself, and figure out when to trigger additional audio DMA from the VBlank interrupt. You know the VBlank runs every 280896 cycles, and you know that the processor runs at 16777216 Hz, so you can do some math to calculate how much data is remaining in the audio DMA stream.
A lot of games simplify the math—it’s easier to start a new audio DMA in your VBlank handler, but that means running at a lower sample rate, which will sound pretty crispy.
YMMV, some people like the crispy aliased audio. If the audio weren’t crispy, the sound designers probably would have adjusted the samples to compensate. Other factors being equal, I’d rather listen to what the original artists heard when they were testing on real hardware, because that is probably closer to what they intended, even though it has a lot of artifacts in it.
> why don’t more games run their audio at 32768 Hz, if that’s such a natural rate to run audio?
I've written some code to play back 8-bit samples (and indeed to wavetable, FM, and VA synthesis) on 8-bit Arduinos using the PWM to output 8-bit audio. That runs at 31373Hz which is a pretty crazy sample rate.
Why?
Because the chip is clocked at 16MHz, and if you program the PWM for no prescaler and "phase correct" PWM where it counts up and back down, so you get a widening pulse in the middle of a "burst", then it counts 510 "steps" of the counter. It's an 8-bit counter so it counts from 0 to 255, then the next step counts back down to 254, and so to 0 again, when the next step takes it to 1.
The reason the nearest neighbour interpolation can sound better is that the aliasing fills the higher frequencies of the audio with a mirror image of the lower frequencies. While humans are less sensitive to higher frequencies, you still expect them to be there, so some people prefer the "fake" detail from aliasing to them just been outright missing in a more accurate sample interpolation.
It's actually the other way round: Aliasing fills the lower frequencies with a mirror image of the higher frequencies. So where do the higher frequencies come from? From the upsampling that happens before the aliasing. _That_ makes the higher frequencies contain (non-mirrored!) copies of the lower frequencies. :-)
Just so that my wrongness isn't there for posterity: This is wrong for a real-valued signal (which is what we're discussing here). I had forgotten about the negative frequencies. So there _is_ a mirror coming from the upsampling. Sorry. :-)
The crispy aliasing of the audio has always felt cozy to me. It’s also a bit of a signature of the system, like the wobbly polygons on PS1. I appreciate that there are ways to change the sound, but it feels a bit rude to label it broken or defective.
I strongly disagree here. I was so hyped for the GBA that I bought it on release day, only to be disappointed later. One of reasons is the lackluster sound; seriously, Nintendo had already built an impressive sound system for the SNES, and then the GBA just had a software-driven DAC? Why did the cheapened out so much?
Same as for the PS1, I always found the wobbly polygons and warping textures painful to watch.
Audio was the thing I could never figure out on my Gameboy emulator. I couldn’t get it to pass basic tests, even without bothering to output sound on the computer.
The loss in high-frequency information is not worth the interpolation. Bass loses its crunch. Percussion fades into the background.
Besides, I personally prefer to play my vgm at the original sample rate, and my soundcard adjusts to the correct rate for each song through fb2k plugins.
I suspect that without nostalgia, the fixed interpolation would absolutely sound better. Unfortunately, nostalgia. The lesson I'm taking away here is that, oh, the terrible resamplings are the aspect of faithful emulation that makes it sound like a GameBoy and not just sawtooths.
I don't quite understand why the author is doing special handling for PSG versus PCM audio.
My GameBoy emulator generates one "audio sample" per clock tick (which is ~1 mhz, so massive 'oversampling'), decimates that signal down to like 100 ksample/sec, then uses a low-pass biquad filter or two to go down to 16 bit / 48 khz and remove beyond-Nyquist frequencies. Doesn't have any of the "muffling" properties this guy is seeing, aside from those literally caused by the low-pass.
>what if, instead of accurately emulating how the GBA PWM hardware works, the emulator uses its own interpolation algorithm to resample from audio channels’ sample rates directly to the emulator’s audio output sample rate?
Then it would be less accurate to the actual console, and thus a worse emulator.
If you combine "GBA Mus Ripper" and "SoundFont MIDI Player", you can get some seriously excellent sound for listening to GBA music.
"GBA Mus Ripper" detects the so-called "Sappy" music driver and extracts and converts the songs to MIDI files, and generates a SF2 soundbank file. Available at https://www.romhacking.net/utilities/881/
"SoundFont MIDI Player" plays back MIDI files. You can configure it to automatically load a SF2 soundbank file in the directory. When you load a converted GBA MIDI file, you get the high music quality of a modern feature-packed MIDI playback engine. Available at https://falcosoft.hu/softwares.html#midiplayer
It's not perfect though, as GBA games do not use true standard MIDIs. Some MIDI controller commands (like modulator wheel) don't translate correctly.
I'm not well-versed in the terms, so I'm not sure which part is the so-called "audio aliasing."
To me, the original has very obvious background noise which the enhanced version removes. But as the author has said, the enhanced version sounds "muffled" (and, IMHO, not just a little), which probably makes most people (including me) feel it sounds worse.
Also, shouldn't most of music be included in the game's official OST? I assume that version would not be limited by the game media's technical limitation at the time and should represent the artistically intended version best.
Edit: apparently in this very case, "Metroid: Zero Mission" doesn't seem to have any official OST release. Unfortunate.
Love what you're doing, but it is funny - I make a lot music in the style of GBA, and specifically bitcrush and downsample to bring in those audio artifacts. They add a lot of high frequencies that give it a great shimmer.
Having said that, there is definitely many use cases where GBA games would want to reduce that artifacting. Keep it up!
The issue is if you pass filter all the high end stuff to try and get rid of that crunchiness, it ends up all sounding muted and muffled. I like the original better (even with the crunchiness).
30 comments
[ 587 ms ] story [ 890 ms ] threadBut this is a bit like those who use smoothing filters. It's ultimately about taste, but it should be recognized that unless the filter is attempting to accurately recreate the original hardware of the era then the original design intent is not being adhered to, and so something may be lost in the "enhancement".
I don't think so, I think you're just getting a high end that isn't in the original audio. In the places where there are high frequencies the aliasing and the hiss just gets in the way.
that drives emotional energy
Seems like a hyperbolic rationalization.
In the mid-1980s the first really affordable sampler was the Ensoniq Mirage, which used the Bob Yannes-designed ES5503 DOC (Digital Oscillator Chip) to generate its waveforms. It played back 8-bit samples and used a fairly simple phase accumulator that didn't do any form of interpolation (I don't count "leftmost neighbour" as interpolation). Particularly when you pitch it down, you get a rough, clanky, gritty "whine" to samples, that the analogue filters didn't necessarily do a lot to remove.
Later on they released the EPS which had 13-bit sampling. Why 13-bit? I don't know, I guess because the Emulator I and II used 8-bit samples but μ-law coding, giving effectively 13-bit equivalent resolution. It also used linear interpolation to smooth the "jumps" between samples, and even if you loaded in and converted a Mirage disk the "graininess" when you pitched things down was gone.
I'm currently writing some code to play back Mirage samples from disk images, and I've actually added a linear interpolator to it. Some things sound better with it, some things sound worse. I think I'll make it a front panel control, so you can turn it on and off as you want.
Then I listened to the accurate version again, and thought "wait, never mind, this one sounds better."
After going back and forth a few times, I think I still agree the original/accurate one is better, but it's pretty close. I really encourage people to listen for themselves.
For what it's worth, I have little to no personal nostalgia for the Game Boy Advance.
I personally do prefer the interpolated versions in most cases because to me the extra high-frequency information just sounds like noise that makes it harder for my brain to process the underlying music. But clearly many feel differently!
It's like saying you can only watch the Simpsons with the exact late 1980s / early 1990s ads that they originally aired with, and everything else is sacrilege.
The "uninterpolated" one is incorrect.
The "interpolated" one is incorrect.
The uninterpolated one has sharp square edges, which isn't correct. The GBA has a 12dB/octave filter at around 12kHz (IIRC) on the output, which the uninterpolated simulated output doesn't appear to have. This would knock the corners off a bit and make it "smoother" and less hissy, but would still have quite crunchy low frequency sounds.
The interpolated one smooths things off excessively, and while it doesn't really have much less spectral energy high up, what's there is in the wrong place.
I’ll add some context here—why don’t more games run their audio at 32768 Hz, if that’s such a natural rate to run audio? The answer lies in how you fill the buffers. In any modern, sensible audio system, you can check how much space is available in the audio buffer and simply fill it. The GBA lacks a mechanism to query this. Instead, what you do is calculate this yourself, and figure out when to trigger additional audio DMA from the VBlank interrupt. You know the VBlank runs every 280896 cycles, and you know that the processor runs at 16777216 Hz, so you can do some math to calculate how much data is remaining in the audio DMA stream.
A lot of games simplify the math—it’s easier to start a new audio DMA in your VBlank handler, but that means running at a lower sample rate, which will sound pretty crispy.
YMMV, some people like the crispy aliased audio. If the audio weren’t crispy, the sound designers probably would have adjusted the samples to compensate. Other factors being equal, I’d rather listen to what the original artists heard when they were testing on real hardware, because that is probably closer to what they intended, even though it has a lot of artifacts in it.
I've written some code to play back 8-bit samples (and indeed to wavetable, FM, and VA synthesis) on 8-bit Arduinos using the PWM to output 8-bit audio. That runs at 31373Hz which is a pretty crazy sample rate.
Why?
Because the chip is clocked at 16MHz, and if you program the PWM for no prescaler and "phase correct" PWM where it counts up and back down, so you get a widening pulse in the middle of a "burst", then it counts 510 "steps" of the counter. It's an 8-bit counter so it counts from 0 to 255, then the next step counts back down to 254, and so to 0 again, when the next step takes it to 1.
And 16000000/510 is 31372.55 ;-)
It's basically doing an accidental and low-quality form of spectral band replication: https://en.wikipedia.org/wiki/Spectral_band_replication which is used in modern codecs.
Same as for the PS1, I always found the wobbly polygons and warping textures painful to watch.
Audio was the thing I could never figure out on my Gameboy emulator. I couldn’t get it to pass basic tests, even without bothering to output sound on the computer.
Besides, I personally prefer to play my vgm at the original sample rate, and my soundcard adjusts to the correct rate for each song through fb2k plugins.
My GameBoy emulator generates one "audio sample" per clock tick (which is ~1 mhz, so massive 'oversampling'), decimates that signal down to like 100 ksample/sec, then uses a low-pass biquad filter or two to go down to 16 bit / 48 khz and remove beyond-Nyquist frequencies. Doesn't have any of the "muffling" properties this guy is seeing, aside from those literally caused by the low-pass.
I absolutely wouldn't. To my ears the second version sounds much worse. Personal opinion etc etc, but wow, it's very very clear to my ears.
Then it would be less accurate to the actual console, and thus a worse emulator.
"GBA Mus Ripper" detects the so-called "Sappy" music driver and extracts and converts the songs to MIDI files, and generates a SF2 soundbank file. Available at https://www.romhacking.net/utilities/881/
"SoundFont MIDI Player" plays back MIDI files. You can configure it to automatically load a SF2 soundbank file in the directory. When you load a converted GBA MIDI file, you get the high music quality of a modern feature-packed MIDI playback engine. Available at https://falcosoft.hu/softwares.html#midiplayer
It's not perfect though, as GBA games do not use true standard MIDIs. Some MIDI controller commands (like modulator wheel) don't translate correctly.
To me, the original has very obvious background noise which the enhanced version removes. But as the author has said, the enhanced version sounds "muffled" (and, IMHO, not just a little), which probably makes most people (including me) feel it sounds worse.
Also, shouldn't most of music be included in the game's official OST? I assume that version would not be limited by the game media's technical limitation at the time and should represent the artistically intended version best.
Edit: apparently in this very case, "Metroid: Zero Mission" doesn't seem to have any official OST release. Unfortunate.
Having said that, there is definitely many use cases where GBA games would want to reduce that artifacting. Keep it up!