Printable View
I do have been working on PCM playback over PSG, but for Master System instead, but unfortunately the results were not all that great, not better than what was achieved in past. Main problem is that it is very quiet compared to the rest, which is also why After Burner has quiet music compared to other games.
In case of MD, PSG is much better off doing what it is supposed to do.
Why are you stuck with low volume for the PSG . . . are you talking about doing playback on just 1 channel?Quote:
Originally Posted by TmEE
The best sounding (though less resource efficient) method would be using 2 or (better) all 3 channels mixed together to approximate higher resolution linear PCM (or perhaps even a ulaw-like scale, or just a custom semi-linear scale -better optimized than sticking to a normal PCM format). The concept is pretty simple (work out optimized values to select for 3 combined channels and build a look-up table to allow translation of PCM samples to 3xPSG values -or, also produce an optimized custom PCM scale that caters specifically to the non-redundant amplitude levels available via the PSG -also limiting it to 1 channel or multiplex/interleave type mixing), but the bigger issue would be establishing the values to use for conversion. (you'd need to know the specific volume steps used by the PSG and then work out an optimized set of the most useful, non-redundant values for PCM playback -and the parameters for that would vary depending on whether you wanted to aim at plain 8-bit linear space or an optimized set of 256 values with no redundant amplitude levels -which would have a dynamic range exceeding that of 8-bit PCM by a wide margin, but with significant gaps left in it too . . . though it would be interesting to see what sort of SNR you could get with that -especially if you threw bitrate out the window and went beyond an 8-bit packed format and used the maximum possible range available with all 3 channels)
For the multi-channel method, there's always going to be some added aliasing due to delay between setting each channel's volume (same issue for playback using the AY/YM2149), but that shouldn't be a major problem. (the YM2149 also has the huge advantage of finer volume steps and 5-bit resolution per-channel)
There's also the issue of the max sample rate possible via the PSG.
You also mentioned before that Afterburner just uses plain 3 kHz 4-bit nonlinear samples (without channel paired mixing/look-up), right? (in which case, it sounds surprisingly good . . . better than a lot of 4-bit linear samples of similar rates I've heard -including the oddly poor examples in PCE ABII- though it could be more an issue with optimization for the recording than anything else)
It's also interesting that it's doing 2 simultaneous channels (maybe the first MD game to do that) with 1 dedicated to music (drums, etc) and the other to speech/SFX. (since you mentioned that all 3 channels are used together -for loudness- and since additive mixing would be impractical, I guess they took the multiplex/interleave mixing route -which should take slightly less overhead than additive mixing anyway)
It's too bad more PSGs didn't use simple linear volume (especially like the Lynx's 8-bit linear resolution ;)) . . . though, another interesting feature would have been manual control over the high/low cycle of the waveform. (allowing actual signed logarithmic PCM playback)
Using more channels will null all good uses for sample playback. I need channels for music and stuff too.
It would still have the advantage of PCM plus all 6 FM channels. (but even then there's the practical disadvantage of taking more overhead than YM playback)Quote:
Originally Posted by TmEE
And there's the whole non-practical demo perspective of what PSG sample playback can sound like at best (which is more like what Joe was asking).
On the practical end of things for single-channel 4-bit nonlinear playback, the poor resolution (and granularity) would probably be most competitive (with 8-bit PCM) at low bitrates . . . so not so much what you'd want for modern homebrew stuff, but a more interesting prospect from the perspective of early MD games (where most samples were limited to around 32 kbps/4kB/s) where 4-bit samples could be 2x the sample rate while the poor amplitude resolution is also less of a problem (as sample rates get higher, that issue becomes far greater).
Albeit, in that case, you'd still have the issue of plain 4-bit PCM to compare that to (played through the DAC), but if the quality was reasonably similar (with similar optimization/preprocessing), using the PSG would still be advantageous when the 6th FM channel was needed. (and, again, the results from afterburner are pretty decent in that regard -well optimized 4-bit PCM should still sound a bit better though . . . but on the other hand, several examples of 8-bit PCM sound worse due to crappy playback on top of low sample rates -perhaps poor conversion/recordings as well- )
And that's short of resorting to actual DPCM/ADPCM/CVSD type compression. (4-bit ulaw-like formats might also be useful for some things -especially speech- while keeping overhead low)
But, again, with modern homebrew, limiting bitrate is not really the main concern. (and a few early games also opted for pretty high bitrates - the 16 kHz 8-bit PCM examples- albeit limited to few/short samples) Well, I suppose making freeware games catering to ROM sizes supported by existing flash carts could still be a factor too.
I was talking in the context of SMS. On MD you would not want to mess with PSG for samples, ever, there just is no point.
Hmmm so what can you get out of the 32X soundwise?
At a minimum? CD quality audio.
http://www.youtube.com/watch?v=ctDCOrE67Ss
Well, not QUITE CD quality, but close. You can only do 9-bit audio at 44100 Hz, so it's not quite there. Most of my stuff on the 32X is 10-bit audio at 22050 Hz. The MOD player used in the yeti demo, for example, is 10-bits at 22050 Hz.
Hmm, that sounds really good then. I would have to hear the original and compare it directly to know any better.
Thanks. The code actually does all mixing at 16 bits for quality, then converts to 10-bit PWM as the last step to preserve as much of that quality as possible. The samples are all 8 bits, but when I scale them for the current volume the channel plays at (plus panning), I scale the samples UP to 16 bits at the same time they get mixed. Having a 16x16=>32 multiply that takes one to three cycles on the SH2 is very handy for audio mixing. The mixing also isn't quite as good as it COULD be - I don't interpolate as I resample the samples for pitch. I might add that later... and maybe some real filtering after mixing for the best sounding output possible.Quote:
Originally Posted by sheath