multichannel rendering

[Tom M.]

I would like to be able to render each voice to a distinct stereo buffer. Currently this doesnt seems to be easily possible?

As far as I understand, one would have to implement a custom Multi_Buffer class holding an array of channel_t[Music_Emu::voice_count()] and then set this custom buffer via Music_Emu::set_buffer(). Am I missing something?

[Michael Pyne]

What you suggest seems to be correct. channel_t is an analog for a specific voice in the Multi_Buffer.

This would only work for music emulators that are derived from Classic_Emu at this point (e.g. the GB emulator, but not the SNES SPC).

Looking at the code I'm not sure that this would work until Music_Emu::mute_voices(0) has been called manually, since the Classic_Emu appears not to use the custom buffer that is passed in except for in mute_voices_, and I've not seen anywhere else where that function might be called as part of initialization.

[Tom M.]

Ok, I'll try to work on this. In case anybody wants to follow: https://github.com/derselbst/game-music-emu/tree/multichannel

But instead implementing another custom class that does pretty much the same as Stereo_Buffer, I'll try to enhance Stereo_Buffer for holding multiple channel_t. By having some

gme_enable_multichannel(Music_Emu* emu, int isEnabled);

Classic_Emu can decide whether to give each voice a separate channel_t or not.

Do you see any chance to get Spc_Emu for multi channel rendering?

[Michael Pyne]

Spc_Emu looks like it would need a bit of surgery within Spc_Dsp to support multi-channel rendering (along with addressing how Echo and FIR would work broken up across independent channels). I don't think it would be horribly complicated though.

[Tom M.]

Ok, so instead of modifying Stereo_Buffer, I fiddled around with Effects_Buffer. It basically works quite well. If multichannel support is requested through gme_new_emu(), gme will allocate an Effects_Buffer that can hold 8 voices (maximum possible). Ofc this has to be supported by the underlying Emu type, and atm only Classic_Emu does. Every other Emu should generate sound as before. The user can test via gme_multi_channel() whether the output retrieved will have each voice rendered to their own stereo channel.

Please review my changes and let me know if I messed smth. up. There is one issue I could not solve, which is that strange stereo mask handling in Effects_Buffer::end_frame().

https://github.com/derselbst/game-music-emu/tree/multichannel

[Michael Pyne]

I'll pulled up the Github page and will try to review it sometime this week. Not blowing you off, it's more a function of (lack of) free time. But I'll try to keep you from waiting too long one way or the other.

[Michael Pyne]

I opened a pull request with your changes and started on reviewing. Not done yet by any means but if you want to take a look I've at least noted something things you could start on.

[Tom M.]

There seems to be a bug when trying to multichannel play vgm files (specifically Sonic 3 & Knuckles). Audio seems to be 4 times faster as usual. I'll look into that, but it may take some time until I do that. Fortunately everything sounds fine without multichannel rendering.

[Tom M.]

Finally had time looking into this. Sonic 3 instantiates Vgm_Emu class which internally uses requires the YM2*** FM sound chip emulators. Those have hardcoded stereo support. There are just so many sample buffers and resample buffers floating around that I completely lost overview. Unfortunately the only option I saw was to disable multichannel support for Vgm_Emu class, see the commit I pushed to my github fork.

[Michael Pyne]

OK, I've applied your commit, hopefully when we get it figured out we can re-enable for YM2*.

[Wohlstand]

Continue at new repo: https://github.com/libgme/game-music-emu/issues/8