Actually this is a long-standing issue. I believe what is happening is that the games are trying to use the OPL synthesizer chip as a 1-bit DAC similar to the PC Speaker. However, this is very timing intensive and DOSBox does not handle it accurately enough.

It works for PC Speaker because DOSBox implemented some sort of buffering system for PC Speaker data.

Another game to test this with is Wizardry 7. If you use PC Speaker for digitized sounds it works fine. If you use Adlib/SB it doesn't (at least, I'd presume it doesn't - I haven't tested for a while because I haven't played with DOSBox lately).

Interesting. I had heard there were some games using OPL chips to act as a DAC to play digitised sounds.

Some mod players and the digital illusions pinball games use it, probably a bunch more.

hmm is this why Xwing doesnt play some samples well?
Recorded a sample from p166 machine with SB16 and compared to dosbox v0.72 (changing laser firing mode)

Also notice the waveform for dosbax doesnt play heaps of the negative part.

.

No, this is a different problem.

ahh ok sorry to barge in on the topic, just say if you want it on its own topic.

This problem has been bugging me for a long time, are there any ideas to fix?

Well it's annoying me as well, but i'm not sure why it happens.
It seems like the "real thing" (opl itself or some other component?)
automatically adjusts the level so the waveform looks fine. I've implemented
something like that for dosbox as well, yet tuning it seems quite hard
(it doesn't really improve the sound as well).
Maybe somebody has/can find information about why this happens.

hmm it looks like the negative part isn't accounted for at all.
Maybe a sign thing

Nope, they select a waveform that has the first part of the sine only
(or a similar one).

Nooo, been waiting for years for them to get around to looking at the OPL-as-DAC issue and you distracted them 😜 hehe

The opl precision issue surely won't be resolved anyways.