Nice idea! That would make those Vibra (pro) CT2502 based cards, like SB16 CT2940, real nice since you can't replace DSP on those.

Edit:
Sorry, I can't resist this. Two days ago I finished modding wavetable header for AWE64 CT4500 (it will be on my next video). That board to fix hanging notes would make a bit nicer lookin mod but more importantly possible to mod CT3600, another great CT2502 based card. I wouldn't even consider modding that one because of the MPU bugs but that board would change things. CT3600 would be like CT4500 but with real Yamaha OPL3 (but that requires a mod as well).

So this fixes both hanging note bugs?

Seems too good to be true 😁

Hahaha, neat idea ~

Love this... hope it makes it to production i'd grab a couple.

matze79 wrote on 2026-08-27, 15:05:

The SB Bug Stomper is the answer to buggy Sound Blaster cards that suffer from hanging notes. Placed between the wavetable modu […]
Show full quote

The attachment photo_2026-08-26_21-11-03.jpg is no longer available

The attachment photo_2026-08-26_21-02-37.jpg is no longer available

The SB Bug Stomper is the answer to buggy Sound Blaster cards that suffer from hanging notes.
Placed between the wavetable module and the Sound Blaster, it tracks all MIDI note events, including Note On and Note Off messages.
If it detects something out of the ordinary, it steps in and stops the stuck note — putting an end to those horrible, never-ending tones.

Prototype Test soon 😀

That's so cool!

Just curious... since it is interacting with Note On\Off messages via the waveblaster header, does it work on other cards? MediaVision Pro Audio Spectrum 16 cards are praised for their output quality compared to SB16 but apparently suffer from hanging note bugs as well. I haven't experienced it personally, but if there is any chance that this mitigates those as well that would be awesome.

It looks like you'd need to solder this to the wavetable card? I feel like if you are going to modify a card it should be the SB since that's where the bug is 🤔

NeoG_ wrote on 2026-08-30, 22:37:

It looks like you'd need to solder this to the wavetable card? I feel like if you are going to modify a card it should be the SB since that's where the bug is 🤔

It looks to me like there will be two headers on the finished board. I would expect this to effectively be "sandwiched" between the buggy host card and the daughterboard, where it will transparently analyse MIDI messages and act as necessary.

I assume there would be a header socket on the finished board.

The main consideration would be the physical offset of connecting a wavetable header. I imagine it will block an adjacent slot, unless in a shared ISA/PCI slot.

That said, this seems brilliant and may finally make my Sound Blaster wavetable headers useful. 😁

I'm interested. How fast can it detect an anomaly? E.g. Is there any delay before the fix kicks in? Basically I'm wondering if playback would be 100% the same when compared to a sound card without the bug.

Hi,

i awaiting prototypes 😀
then we will see.

You put the Stomper on your SB card, then connect your Wavetable to the Stomper.

It simply intercepts the communication and tries to identify stuck notes. It tracks all notes and looks for “high-pitch” stuck notes and other anomalies.
It adds miniscule non audible delay of around ~2ms to Midi. The MCU is pretty fast.

@Dreamblaster’s new Wavetable also implements this idea.

On my buggy SB Vibra 16, it actually works. 😀

Great little project. Have the prototypes arrived?

Yeah 😀

Very exciting! Found any issues with the prototype yet?

It works as expected.

What works so far:

1. Automatically detects hanging notes — in two ways: A note that plays for an unusually long time is automatically turned off. The timeout depends on the pitch, since the bug usually affects higher notes: ~1 second for very high notes, ~3 seconds in the mid range, and ~10 seconds for low notes, where legitimately long notes such as bass notes or pads are normal. If the same note is triggered twice in a row without being properly released in between, it intervenes immediately without waiting for the timeout.
2. Recognizes reset messages from General MIDI, Roland GS/MT-32, and Yamaha XG, and clears its internal state so that nothing remains stuck from the previous game.
3. MT-32 instrument remapping from GM → MT-32.
   Note: SAM-based wavetable modules such as the Dreamblaster S2 can already handle this themselves.

Yet untested / Experimental

1. Effect: Delay Notes / Echo Mode
2. Effect: Add Reverb

Test setup: McFly (MT-32 mapping + SoftMPU)

It is a cool idea. And the logo is even better

I noted back then that a game's sound driver can fix-up SB16 midi bugs like this:
Doom 'MBF' for DOS, Maintenance release 2.04

A technical note on the possible lockup with the Sound Blaster 16 cards, when using digital effects together with the build in MPU-401 Midi interface.
This is a problem also found in other games. As discussed earlier the patch SBMPU401.EXE (inside waveptch.exe) offered by Creative prevents such lockup.
Basically, what "SBMPU401 /E" does is is write 0Bh to undocumented mixer register 83h. "SBMPU401 /D" reverses this by writing 0Fh to the same register. My tests indicate that the /E register tweak prevents unexpected interrupts on the SB IRQ; these are ones that contain the MPU tag (100b). Digital sound interrupts keep coming, as they are necessary. The distinction of interrupt types can be made because SB16 introduced mixer register 82h, which informs on the type of interrupt that was given. But AFAIK the MPU should not send interrupts in UART mode to begin with. MBF now filters out and acknowledges any interrupts not tagged 'digital effect 8/16bit' and that works well. SBMPU401 is no longer needed for this game.

Alleg30B\src\djgpp\sb.c

1   // GB 2017 This is an attempt to fix the situation when using a SB-16 together with its buggy MPU-401 interface.
2   // See also "mpu_output" in mpu.c. 
3   if (sb_hw_dsp_ver >= 0x400) {           // DSP 4.xx only 
4     outportb(_sb_port+4, 0x82);           // SB mixer interrupt status register (82h)
5     if ((inportb(_sb_port+5) & 3) == 0) { // Not a 8- or 16-bit DSP interrupt, but maybe mpu-401 instead?
6	   _eoi(_sb_irq);
7       return -1;                          // This interrupt was either bugged/false or not for this handler, exit now.
8	 }
9   }// End of fix. 

And:

1   // GB 2017, on SB16: increase buffer size -> minimize DMA interrupts -> less MPU interrupt problems.
2   if (sb_hw_dsp_ver >= 0x400) while (sb_dma_size < 2048) { sb_dma_mix_size *= 2; sb_dma_size *= 2; }
3   // GB 2017, DSP 4.11, 4.12 or 4.13 are affected by hanging notes after digitalsound interrupt.
4   if ((sb_hw_dsp_ver >= 0x40B) && (sb_hw_dsp_ver <= 0x40D)) 
5      { sb_dma_mix_size *= 2; sb_dma_size *= 2; } // 4096 buffer, problems gone :)

Alleg30B\src\djgpp\mpu.c

1// mpu_output: Writes a byte to the MPU-401 midi interface. Modified GB 24-8-2017
2static void mpu_output(unsigned char data)
3{
4   int timeout;
5
6   if (wait_for_mpu(0x40, _mpu_port+1))   
7   if (_sb_irq>0)                             // If timed out: Digital section active?
8   {
9	 _enable_irq(_sb_irq);                    // GB 2017, to prevent crash on older models of SB16, like CT1740 and CT2290. 
10     for (timeout=0; timeout<99; timeout++);  // Allow SB16 digital section to service interrupt if necessary
11   }                                           
12   outportb(_mpu_port, data);
13}
14static END_OF_FUNCTION(mpu_output);

But I switched to Crystal CS4232 based soundcards later, so it has been a while since I actually used a SB16/AWE cards.

Neat, though I dread the idea of people paying actual money for Vibra cards thanks to this. What are you using to monitor the MIDI stream? If there's room a nice bonus feature would be reverse stereo, for people with Audician 32s etc who don't want to have to cut traces for proper wavetable sound.

terrorinstinct wrote on 2026-09-29, 04:17:

Neat, though I dread the idea of people paying actual money for Vibra cards thanks to this. What are you using to monitor the MIDI stream? If there's room a nice bonus feature would be reverse stereo, for people with Audician 32s etc who don't want to have to cut traces for proper wavetable sound.

Is it miswired L/R, or is it an issue with the MPU interface?
It might be possible to fix it by using the MIDI Pan controller:
10 Pan (MSB) 0–127 Controls the left and right balance, generally for mono patches. A value of 64 equals the center.

But for now, I'm still working on the firmware.

I'm capturing the MIDI data from the Vibra and comparing it with the original MIDI data.
That should also help me come up with a better filter.

The stream is captured by using python3-serial and ftdi adapter.

I checked the options for this, and it would be best to have jumpers for channel switching.
This would require a new PCB to support it.

---

Note tracking and sending Note Off messages are now only a fallback option if filtering does not work.

Sound Blaster CT2960 Vibra (HP) with DreamBlaster S2 only:
https://rumble.com/v7gbhba-soundblaster-vibra … ug-stomper.html
(One channel was not recorded because the cable is flaky, but you can hear the bug 😀

Sound Blaster CT2960 Vibra (HP) with Bug Stomper and DreamBlaster S2:
https://rumble.com/v7gbhey-soundblaster-vibra … blaster-s2.html

Updated Entry #1.