#1248 closed defect (fixed)
SAM: some MIDI channels missing on MT-32
| Reported by: | SF/logicdeluxe | Owned by: | athrxx |
|---|---|---|---|
| Priority: | low | Component: | Engine: SCUMM |
| Version: | Keywords: | ||
| Cc: | Game: | Sam and Max |
Description
ScummVM 0.5.3cvs (Sep 26 2003 10:12:25) I compared with the original interpreter and it sounds quite complete to me there. ScummVM lefts out some channels at certain situations. The channel mapping is correct with the recent MT-32 patch, so they aren't played on a unsupported channel either, as I can confirm on my MU5 display. It probably has to do with iMuse priority settings. The problem is very noticable in the teaser before the opening credits for instance.
Ticket imported from: #812998. Ticket imported from: bugs/1248.
Attachments (3)
Change History (23)
comment:1 by , 21 years ago
| Owner: | set to |
|---|
comment:2 by , 21 years ago
comment:3 by , 21 years ago
Status is "still trying to formulate a way to test theories given that I don't have the hardware in question". :)
comment:4 by , 21 years ago
| Priority: | normal → low |
|---|
comment:5 by , 21 years ago
I can hear the issue in the opening cutscene. The problem is (1) the music for that cutscene uses 14 non-percussion channels, (2) only 2 of those channels have priority overrides, and (3) the set of channels that could be considered "core instruments" are scattered all over. (The third point makes it hard to allocate channels based on, say, whether a part falls in the Base MIDI or Extended MIDI range.)
Since there are 14 channels in use, the original distro MUST be discarding some of them for the MT32, but probably only the more background instruments. Is it possible for you to record the output of the MT32 when running the original distro? That way I can try to isolate which instruments it's discarding, and from there determine how they made their selection.
I'm lowering priority, in the meantime. This doesn't seem to affect a lot of the game, since more tracks appear to stay within the 8-channel limit of the MT-32.
comment:6 by , 21 years ago
I would record the original output, but I don't have a sequencer in order to do that. Is there a kind of sequencer I just could add to the Windows MIDI mapper? With that I could run the game in DOSBox and record its MIDI output. I searched the internet for such a tool but I didn't found such a thing yet. I'll keep trying. And I'll post the results here as soon as I find a way.
comment:7 by , 21 years ago
I mean the audio output. I want to hear exactly what you're hearing when you drive the MT-32 from the original Sam & Max distro.
comment:8 by , 21 years ago
I recorded it running in DOSBox. (Just installed the floppy version which was the last real mode incarnation of SCUMM afaik.) It runs a bit slow there, hence some passages in the music last slightly longer than usual. I hope the sample is long enough for you.
by , 21 years ago
| Attachment: | samintro.ogg added |
|---|
by , 21 years ago
| Attachment: | sam-midi.zip added |
|---|
MT-32 vs. GM recorded from the original interpreter
comment:9 by , 21 years ago
So I finaly managed to record the MIDI output from the game. I made files including the entire intro from the initializing of the game up to the point you get control over Sam. So you can easily load them in a MIDI editor and compare every single event. Don't take the timing too serious! The ticks are just counted in emulated VGA retraces of DOSBox, which is 60 Hz, I guess. ;-) I hope those MIDI files are of any help for you.
comment:10 by , 21 years ago
Whoa, this is good stuff! Your DosBox mod looks quite useful. It'll take me a while to figure out this problem with it, but I already noticed something unrelated. I saw the pitch bend range setup you mentioned, in the init sequence. So they set everything up to +/- 16 half-steps, and they do it in the Patch non-temp area. And they do it for all 128 patchs -- yikes!
So I'm interested in seeing how the original distro handles changes to the pitch bend range mid-game. This happens for the music outside the Mystery Vortex -- an override of +/- 12 half-steps is applied to the channel doing the eerie sliding flute sound.
Marek: Do you think you could generate one of these cool DosBox MIDI dumps for that music? Basically start capturing from the USA map and stop a few seconds into the Mystery Vortex music. The pitch bend override is one of the first events of the track.
That dump may very well tell us what type of SysEx call we should be making in ScummVM.
comment:11 by , 21 years ago
> So I'm interested in seeing how the original distro handles changes to the pitch bend range mid-game.
It doesn't. At least not on the hardware side. All MIDI games since Monkey2 did setup the hardware to do +-16 half steps (as you can see in the MIDI files). iMuse must convert them internally, similar to the way ScummVM currently does with MT-32 output. ;-) V6 games use the same approach with GM, exept that the pitchbend setup is done per channel instead of per patch, but also there the setup is never changed during the game. Funny, just as I'm thinking about it, I can't remember ever hearing a pitchbend in MonkeyEGA/VGA and LoomEGA as threre is no such initialization. Maybe Michael Land didn't had a pitch bend wheel back then? I'll check the Vortex. If this is a problem, we have to implement the whole thing in ScummVM and convert pitchbends to +-16 like the original does it. MIDI dumps for MT-32 and GM uploaded. I traveled from the map to the vortex and after a few seconds back to the map and then quit the game. (The silence at the beginning is the time for copy protection screen and loading of the savegame.)
BTW., will I be added to the credits as MIDI tester and contributor?
comment:12 by , 21 years ago
Now I tested the Vortex with current ScummVM: There is no pitch bend problem with the way it is currently implemented. My MT-32 plays the Vortex just fine. If there is a pitch bend of more than +-12 half steps anywhere in any game, then it would be a problem with the currently implemented MT-32 code, but I am not aware of such a cue so far. If I find one, I will post this.
comment:13 by , 21 years ago
Well, yeah, we patched ScummVM to do internal pitchbend scaling and send out pitch bend values that assume the MT- 32 resets itself to a pitchbend range of +/- 12 half steps.
Looking at the GM init sequence for S&M, I see that the original distro also set up GM devices for +/- 16 half steps, which means it must have just done internal scaling of pitchbend values for EVERYTHING. I suppose ScummVM should take the same approach, then.
BTW, you are correct -- I think MI2 is the earliest game where I can remember hearing pitch bends. That may be because of the large number of messages required to conduct a smooth-sounding pitch bend. Earlier machines might not have been able to process a high frequency of events without lagging.
> BTW., will I be added to the credits as MIDI tester and contributor?
That's not my call to make. We really do appreciate your many contributions to the project; however, keep in mind that there are many such people who make frequent contributions, and we cannot list them all by name. I hope recognition is not the primary motivation for your contributions; remember that this is an OSS project, and getting billed in the credits doesn't really amount to much.
If it makes you feel better, know that I jump for joy every time I see a new bug report from you in my Inbox. ;)
comment:14 by , 21 years ago
Since there is a bug in several wide spreaded Dream chipsets which prevent pitch bend depth setups of more then +/- 12 half steps from working correctly, it would be a good idea to not implement it EXACTLY the way LEC did. If there really is no cue using more than +/- 12, and I guess so, there is really no good reason not using 12 instead of 16. This would remain the MT-32 pitchbend setup unnessecary, it is kind to those bugged wavetables, and still would perfectly work with all other GM compilant devices. We could add a warning whenever a pitch bend depth is set to more than that. Let's see if this ever occours! So far the deepest I found is 12 at SAM's Vortex, DOTT's LEC-Logo and the crashing bird in the intro right after that. If there is a pitch bend depth of more than +/- 12 after all, I would prefer not to change the current behavior, since it even works very well with those bugged wavetable chips, which is not the case with the original interpreters. After all, ScummVM is always proud to do things better, isn't it?
comment:16 by , 20 years ago
Yo folks, could somebody tell me the status of this item? In particular, is the original issue (some channels being left out) still there?
The pitchbend stuff sounds very interesting, but maybe it should be in a separate tracker item?
comment:18 by , 19 years ago
| Owner: | removed |
|---|---|
| Status: | new → closed |
comment:19 by , 19 years ago
Setting to pending, and unassigning from Jamieson (as he isn't really working on ScummVM anymore, unfortunately :-/)
comment:20 by , 2 years ago
| Owner: | set to |
|---|---|
| Resolution: | → fixed |
I have made a fix. Thank you for your patience ;-)
What is the status of this item?