I would test outside the machine on the bench. Does speech lock up?

Maybe U8 or it's socket, also check r9

Do mean the U8 the speech chip when you say processor?

Do you have -4.5 to - 5.0 vdc at test point 4 ? If the voltage is low it can lock U8. The -5 volt regulator must be steady at the correct voltage.

Which version of S&T do you have?
Early version or late version?

As a matter of fact the TMS5200 is a voice synthesis "processor".

I still think it is the three transistors that control the -5 volts
that is located on the upper right-hand side.

The negative 5v is made by this circuit. If the -5v is stable and minimal ripple, the circuit is OK. The board will not POST OK the speech chip if -5v is missing, probably not your issue.

negative 5v (resized).png

Note the jumpers about volume control.
st jumpers (resized).png

Yes, there are 2 Bookkeeping registers that are part of the "digital" volume control feature that usually isn't used since the board are typically jumpered to use the two trim pots on the board. It kind of makes sense that at the end of the boot up sequence (or maybe game start) that the Main MPU board sends the volume settings digitally to the S&T board. I always make sure to set registers 16 - 21 to 03 just to be sure. Of course, the volume ones go up to 15.

Did you try the S&T EPROMs on another working board? Did you try another set on this board? You didn't say if you swapped the TMS5200 between boards either?

Have you replaced chip sockets and header pins ? It's amazing how unstable these old boards can be until you replace those.

Have you checked the power-up reset-circuit ? There's a capacitor (C1) that could be gone and causing the timing to go wild.

Why is one not showing the -5 volt timing circuit schematic to the TMS5200?

The speech chip oscillator circuit? It is set by rough value components, but not specific timing is not critical. It being out of spec should not cause a low volume issue. Fast/slow pitch and or talking speed.

The "speech power sequencer" circuit is just an extended delay to turn on the -5v connection to speech chip after the main CPU's reset pulls high. Since there is any speech present, it can be assumed to work.

Yes R9 is important!

Yes, the timing of the TMS5200 (internal) oscillator is maybe not supercritical but has to be good.

I spent some 3 months trying to solve a problem with occasional garbled speech on one board only to finally discover that the oscillator frequency of the TMS chip was slightly off due to a bad R9. In my case the board usually worked but once in a while the speech was rubbish and much shorter duration. The TMS5200 internal oscillator affects also the timing towards the CPU so funny things can happen.

I always measure the frequency of TP11 and adjust R9 to whatever value is needed to get precisely 160kHz (6.25 us) at TP11. On my board an R9 of 124 kOhm was needed. Schematic states 130kOhm while the actual measured original R9 value was 150kOhm!

With that small adjustment my board is rock solid and the speech crystal clear. Maybe it does not solve your problem but could be worth a check if not already done. An oscillator at correct frequency is always a good starting point.