Are you running the normal game roms in U26 and U27 when you have the sound test rom installed at U21? The game roms should be installed in U26 an U27 when running the sound test. However, you need to take the game sound rom out of U22 when you run the test.

The sound input signals at U9 pins 2 through 9 should not be active with the game roms installed. If the sound test rom is running, you should pulses going up and down on pins 10 through 17 of U9. They won't be fast pulses, it will cycle back and forth about every half second.

It sounds like the sound section of your board is not booting. You can try to force it to reset by briefly grounding pin 40 of U24. If this works and you see strobing on pins 10 through 17 of U9, you know that the reset signal is not getting to the processor when the board boots. If it doesn't work, I would start working my way through the rest of the Leon's article on sound issues.

What voltages are you providing to the board on the bench?

U32 is a 74ls08, and consist of AND gates. You have to have high inputs at the 1 and 2 to have a high output at 3. However, I don't think U32 would keep the test rom from running.

You are correct, if that IRQ signal you posted came off of pin 4 of U24, that doesn't look right. I'd check the 1k SR upstream of it for cracks. It should be holding the signal high.

I asked about the voltages because of the speaker on your work bench. It won't work unless you are providing +12v and -12v as well +5v. The logic seciton of the sound chip should still work they way you have it setup. At this point, I would pull the roms off the board completely, and then check the data and address lines. Without the game roms installed, the CPU should start executing it's non operative instructions looking for a program to run, and all the data and address lines will strobe. If you don't see any activity at the CPU, then the processor may be bad. You can also check the data line and address lines for shorts that cause one of them to hang low. If the CPU, data lines and address lines all check out, then I would go looking at the chips that control the chip select (from memory, I think they are 139s).