Thinking it's a spike or noise on power up, I methodically jumped a 4700 µF cap to each voltage on the sound & speech power supply... one side to ground and the other to power... and this made no difference. Same sound.
Then I visually inspected the big caps in the cabinet. Besides the "big orange" on the 38 VDC (which is not orange - had already been replaced at one point), there is another called C1 on the 12 VDC rectifier. This one has definitely been replaced at some point, but it looks pretty old. It was too big for the original mounting clip so you know it's probably something even older that somebody pulled off a shelf or recycled it from who knows where. This is the 12 VDC that goes into the main PS in the lightbox... the same 12 VDC that is used to generate 5 VDC for logic and processors including the sound & speech board circuits. EDIT: As per PinWiki, the 12 VDC capacitor is a.k.a. "big orange". I was confused because my C1 is not in same position on the board as the PinWiki photo.
As a test, I jumped in a cap from 5 VDC to ground and I still have the issue, but now it's changed into a completely different sound. Remove the cap and the next time it happens, it's back to the original problem sound.
I think I'm finally on the right track and here's my latest working theory:
I already know the sound is being generated from the speech circuit. The 12 VDC that powers the SC01 voice chip is coming from the sound & speech power supply - ruled that out. The 5 VDC, however is coming from the main PS, which is generating it from the 12 VDC that is being conditioned by questionable C1 in the lower cabinet. This is the same 5 VDC that is used to generate the various spoken words. I don't think it can be coming in through the U13 logic chip since the processor hasn't fully booted, this spike must be coming straight through the RP1 DIP resistor package and the SC01 voice chip is converting it straight into speech - which actually sounds like a long deep groan. Ironic. LOL.
Again, once powered up, sound & speech operates perfectly fine, so another reason to rule out any of the logic chips.
For sure I am going to replace C1 in the cabinet and probably the other two or three caps just to be sure.
Like I said before, what I don't understand is why this wouldn't be a more common issue with all of the sound & speech machines before these capacitors are replaced.
EDIT: As per Clay's Guide:
This is where the designer of this board was seriously off his face. It’s really quite clever but unnecessarily complex compared to what modern machines use.
This circuit all revolves around the now defunct SC01A chip (U14). This chip produces phonemes (I had to look this word up)- basic units of speech. Vary the clock frequency to it (Pins 15 & 16) and it speaks- Wow! It can also produce sound effects the same way.
See the diagram where I’ve outlined the sections of the board and look at the speech output section. Basically two latches (7475 – U11 and U18- controlled by U10) receive data from the MPU , pass it through a DAC (U19) which converts the signal into a proportional current. This then passes through a frequency adjust LM741 (U21), a couple of transistors (Q1 and Q2) then into the speech chip (U14) at pins 15,16. This is the variable clock signal for the speech chip. The pot R16 controls the speech volume. The R6 is a "frequency adjust" pot, and R13 is an "amplitude adjust" pot.
Data is also sent to the speech chip from MPU to the Level Shift 74LS05 (U13), pulled up by a Dip resistor (RP1) and into pins 9 to 14 of the speech chip.
Voice and effects pitch is varied from the MPU by making Pin 13 of Latch 7475 (U9) high then transferring the data to U12 then onto Pins 2 & 3 of the speech chip(U14).