I have for quite some time been trying to get my FG Squawk & Talk board (AS-2518-61A) to talk properly and finally, it works! I thought I’d put my findings here. Maybe someone will find this useful.

The S&T board produced occasional garbled phrases or words. It could happen at the beginning, middle or end of the phrase or a complete phrase was just rubbish. It happened maybe 10% of the time. The rest of the time the voice was loud and clear.

I searched the web extensively in the hope of finding a solution, but did not find anything. The closest I got was this topic on pinside: “squawk talk occasional garbled voice problem”.
It describes the same problem that I had. Unfortunately there is no final conclusion in that thread.

Well, to cut it short, after having gone through the entire board, new caps, test with all new ICs (PROM, PIA, CPU, TMS5200NL) etc. the problem was still there. As a last try, I checked the resistor R9. According to the schematic and the color codes on the resistor, it should have been 130 kOhm, but when measured it was 150 kOhm!

After replacing the (faulty) resistor with a good one (measured to be 133 kOhm), my FG now talks very clear!

Some technical background info:
====================
Before finding the faulty resistor, I tried replacing the U8 (TMS5200NL) with a new device borrowed from a friend. After this replacement the board worked very well, so I thought the problem was solved. I therefore ordered a new U8.

After getting the new U8, I installed it and expected everything to be singing and dancing, well at least talking ... Great disappointment! There were more garbled phrases than ever before!

So, the conclusion of this must be that there is a spread in some internal parameters between each U8 device causing a slightly different behavior. In fact, the speech synthesizer U8 (TMS5200NL) uses an external resistor (R9) to set/trim the internal clock frequency of the device. This allows to compensate for slight variations in the parameters between different TMS5200NL devices.

I should probably from the very beginning have measured the frequency at TP11 (TMS5200 clock). This clock should have a frequency of 160 kHz (for a speech sample rate of 8 kHz). If this frequency is not correct, there are only two possibilities, either the TMS5200NL is bad or the (real) value of R9 is not matched to the actual TMS5200NL device used.

Considering the huge price difference between a resistor and a new TMS5200NL, checking and trimming the value of R9 to get a good clock frequency out of U8 should probably be the first thing to do whenever the S&T board produces any kind of strange speech.

Even though 130 kOhm might have been a good value of R9 in the early 80’s, the parameters of the still existing TMS5200NL might have drifted over the past 30-35 years, so also another value can possibly give better performance.

Thanks for your reactions. I am completely new on Pinside and this is my very first topic here so its extra fun to follow the discussions.

I have been fighting with the S&T the last several months and it would be great to be able to focus on something else again..... but before moving on, I would indeed also be very interested to know if the trick with the R9 solves the problem also for any of you, including the case where always the same phrase is garbled.

Thanks for all your reactions!

Yes, the internal operating frequency of the TMS5200 is defined by the value of R9. The input pin of the TMS where R9 is connected has a very high impedance, which makes it very sensitive to noise. I believe thats the reason for capacitor C51 across R9, i.e. to filter out stray noise. The actual frequency of the TMS is thus probably entirely defined by the TMS parameters for a given resistor value. This could explain the fact that different TMS devices talk differently. If the value of R9 is kept the same, dropping in different TMS devices will cause each device to be running at slightly different frequencies.

In order to achieve an operating frequency of 160kHz, the value of R9 could be between 120kOhm and up to 200kOhm! There is thus a very large spread in the parameters.

I just measured the clock at TP11 and got 6.6us clock period (using R9=133kOhm). Just to see the effect, I then temporarily added 1.8MOhm in parallel to that which gives an R9=124 kOhm. With that value I got almost spot on a clock period of 6.25us (which is exactly what the TMS5200 is expecting). Will probably replace the 133kOhm I have now with a 124kOhm.

Originally in the 80's I think that indeed the parameter spread of the TMS devices was less pronounced and Bally probably bought a whole bunch of parts manufactured roughly at the same time. For Bally it was thus good enough to fix R9 to 130 kOhm. After all it does not matter too much if different games talk slightly different as long as the oscillator frequency is sufficiently good to avoid the garbling.

Would be really interesting to see if both the occasional and consistent garbling can be recreated/resolved by the trimming operation.

I believe the TMS is a purely digital device with digital filtering etc internally and the only analogue parts are the oscillator and the digital to analogue converter at the output. There could be internal (digital) failures due to the age of the devices but I would expect them to be rare. Analogue parameters are a different story. I would think they can easily drift a lot over 30 years and could impact a lot the resulting clock frequency.

Although I have not made extensive research, I would in fact expect that it should be possible to get different devices to talk reasonably similar by means of trimming R9.

If you have TMS5220 devices available, I would do the same test I did today. Try trimming the clock period to 6.25us. I think there is a good chance you will get it talking properly unless there is a difference in the internal coding tables between the TMS5200 and the 5220.

==========
Considering the problem with the occasionally garbled phrases, my hypothesis is as follows:
* In my case the value of R9 was too high => clock frequency too low. This was also clear from the pitch and speed of the speech which was lower.
* As a consequence of the (too) low clock frequency, the TMS device runs slower overall. Normally, the TMS expects that the CPU waits 10us after a write before the next write occurs. If the TMS runs slower, this could also imply that the TMS will expect also a longer wait time from the CPU.
* If the CPU then occasionally violates this longer wait time, the effect could be an occasional lost byte of speech data.

This is pure speculation so far from my side so I could be completely wrong. I might try to check if this theory fits sometime in the future. Any feedback would be much appreciated!

Yes your R9 looks a bit worn.

Looking at your board, I believe you have the first generation S&T, AS-2518-61 with an R9 of 240kOhm (which is in line with the schematic I have here for that board). I have no clue why these early boards had 240kOhm but its quite different compared to the 130kOhm on the A version.

You also do not have the C51 that is present on the A version (but thats less of a concern I think). You can always add one later if that seems to be necessary (to reduce any noise).

If you have the possibility, hooking the board up and measure the current clock period or frequency at TP11 could also be very interesting.

Curious to see what you find!

Another interesting observation is that you have a newer TMS than I have, but an older -61 board. My date code is 8048 while yours is 8109.

An easy test to check if the value of R9 is part of the problem could be to add another resistor in parallel to the one on the board to bring the resulting value of R9 down to around 130kOhm which is on the lower end of the trimming scale (and in line with newer boards). Has to be done with care of course but no need to solder. I used two crocodile clips (or whatever they are called) to temporarily attach an additional resistor.

Yes, there was a change of R9 between -61 and -61A. The -61A is included in the Fathom schematic on IPDB. The change record says: "R9 was 240K". Pity that it does not say why!

Have now installed R9=124 kOhm which is the matching R9 for my 5200 => TP11 clock period of 6.25 us.
Done a lot of self tests and triggered speech generation in various situations and played many games. The board seems to be rock solid now, no garbling at all. I just hope it stays like that...