Speak and Maths Circuit Bend - Pitch Control Ribbon

The vintage Texas Instruments Speak & Spell and Speak & Math(s) attract the circuit bending community who enhance these devices using an experimental technique of pin probing and adding passive electronic components to create novel sound effects.

A standard rotary potentiometer is commonly used to modulate the pitch of the spoken voice on these educational toys. This article shows the addition of a ribbon controller to a Speak & Maths to provide a pitch bend feature. A Spectra Symbol ThinPot was used for the ribbon controller together with three resistors to set the main oscillator frequency range - this generates a clock which ultimately sets the pitch of the voice.

This hardware addition was done as part of the cleaning and repairing of a Speak & Maths to turn it into an even more interesting, truly unique Christmas gift.

1. A working Speak & Math(s) or similar.
2. Spectra Symbol ThinPot (SoftPot) 100mm 10k (solder tabs is ideal, 1% or 3% tolerance): Digi-Key (3%) | Mouser (1%)
3. Three wires, ideally red, black and another colour.
4. A 1k, 4.7k and 20k resistor.
5. Soldering iron and solder.
6. Heatshrink and something to shrink it.
7. Something to make a rectangular hole in the plastic case ideally at an angle.
8. A Dremel-like drill with a 1.5mm bit and 1mm bit and a small rectangular needle file was used.
9. Good quality scissors to slightly trim the edges of the ThinPot's tail.
10. Epoxy resin glue to secure the end of the ThinPot.
11. Craft knife to score plastic and trim away any swarf.

The chip in the Speak & Maths that recreates speech from data in the ROMs is a 28 pin DIP CD2801NL (F delta 8122). This has a secondary role producing two different clocks for the main microcontroller and ROMs.

The pinout for a another version of this chip, the TMC0280, are listed below based on furrtek.org site.

The Speak & Maths powered at 6.0V was turned on and the levels/signals were checked with an oscilloscope with it idle, descriptions are shown below.

The signals looked like plausible matches for the aforementioned pin names.

The Texas Instruments TMS5100 is widely described as being equivalent to the TMC0280/CD2801. The description of the pins from the datasheet aligns with the previous description and check and is reproduced below.

The TI TMS5220C looks fairly similar but has a different pinout. The internal oscillator on both has a nominal frequency of 640kHz. Both datasheets are attached to this page.

The TMS5100 datasheet doesn't describe how to change the frequency of the internal oscillator but the TMS5220C datasheet does show a method to change the frequency, the figure is shown above and the description from the datasheet follows.

An external resistor (Figure 4-1(a)) sets the internal oscillator frequency. The oscillator can be adjusted to correspond to the sampling frequency that was used when the speech was encoded. Use of a shunt capacitor is recommended to prevent circuit layout and environmental noise from affecting device operation. To adjust the oscillator, monitor the frequency at ROMCLK. This buffered output, which is one-fourth the oscillator frequency, is not affected by the input capacitance of the measurement equipment.

The emphasis on a buffered output is interesting. A basic oscilloscope probe can load and potentially affect a signal output at MHz and above frequencies.

It's possible the CD2801 (TMS5100) has an undocumented similar feature for adjusting the frequency.

Clock-like waveforms were observed on several of the pins on the CD2801. Four oscilloscope traces relative to the negative terminal of the battery pack (0V) are shown above**.

1. ROMCLK (pin 3) - 158kHz square wave with duty cycle around 50%.
2. CPUCLK (pin 4) - 318kHz square wave with duty cycle around 50%
3. CROSC (pin 6) - 481kHz sine wave.
4. RCOSC (pin 7) - 437kHz sawtooth wave.

The waves are far from perfect and the CROSC and RCOSC do not cover anything near the supply voltage range. The frequency of CROSC and RCOSC are probably being affected by the passive probe of the oscilloscope and are likely to really be running at the faster rate of 636kHz. The CPUCLK will be this clock divided by two, the ROMCLK will be a further division by two.

Some experimentation demonstrated that adding resistors or capacitors to RCOSC (pin 7) from VSS (6V) or VDD (-3.5V) alters the internal oscillator rate significantly and can even stop the oscillation. This pin looks useful to vary the oscillation rate but using this is tricky in the absence of a description from a datasheet or other reliable source. Some caution in keeping the currents very low would be prudent here to reduce risk of any damage to the chip. Variable resistors in the form of potentiometer are more common and more compact than variable capacitors making them a better choice.

For a different TI chip, the TMS5220C, the OSC pin can used with a 100k potentiometer to VDD to adjust the internal oscillator frequency. This pin has an alternate function as an oscillator input when an external oscillator is selected.

** The frequency counting isn't working well here making the Fnct values obviously incorrect.

The Spectra Symbol SoftPot is a linear position sensor which works with gentle finger pressure. There are two important differences between this linear potentiometer and a standard rotary potentiometer.

1. The wiper (called a collector in the datasheet) is not connected to anything if no actuation force is applied to the sensor meaning there's near infinite resistance if the sensor isn't being pressed.
2. Multiple simultaneous presses are possible and can be problematic.

This means the SoftPot should not be used as a drop-in replacement for a potentiometer without consideration of these two features.

1. If the SoftPot is connected to an analogue input on a microcontroller and the sensor isn't being pressed this input will float and have an unpredictable voltage. In this case a high value external (or internal in the microcontroller) pull-up or pull-down resistor is one solution.
2. When used as a potential divider across a power rail to ground the worst case scenario is two firm presses at either end of the SoftPot. Theoretically, this creates a zero resistance path and will short the power supply. The actual resistance is more likely to be 50-200 ohms but this can't be guaranteed and isn't ideal. A 1k serial resistor or higher can be used with the SoftPot to safely cap the short circuit current in this scenario. This will shift and reduce the voltage range of the potential divider.

The SoftPot is self-adhesive. The wiper portion of the sensor and the end with the three terminals (called the tail in datasheet) bend but both want to revert to a more-or-less flat state. A tight bend or crease might damage the conductors inside the SoftPot. The datasheet doesn't currently discuss the minimum bend radius**.

The ThinPot as the name suggests is just a more compact version of the SoftPot.

** Contacting the manufacturer for more information is an option here.

The pitch can be controlled on a Speak & Maths with the simple addition of a 10k Spectra Symbol ThinPot and three extra resistors, 4.7k, 20k & 1k. The circuit is shown above as an animation in the Falstad Circuit Simulator. The finger presses are modelled as closing a single switch. A double press is also shown with two switches closed and the current can be seen to increase. The RCOSC oscillator signal isn't modelled on pin 7.

The resistor values were found by some experimentation (see below) and are good values in terms of

1. preventing multiple presses on the ribbon from causing any near shorts on the power and
2. keeping the current low for sinking/sourcing on RCOSC (pin 7) of the CD2801.

These values were determined with a rotary potentiometer on a small breadboard which will have a few picofarads of parasitic capacitance. The RCOSC pin is extremely sensitive to capacitance even touching the pin with a finger will change the oscillator frequency. This makes it very likely the final implementation will have a slightly difference frequency range.

Increasing the CPUCLK above 320kHz is a form of over-clocking. As the frequency is increased at some point one of the chips will stop working perhaps in an erratic manner!

** The values shown going down to 50kHz caused the Speak & Maths to turn off as the frequency dropped below around 50kHz, presumably due to the clock generation stopping for some reason.

A Spectra Symbol 100mm 10k ThinPot 3% tolerance with solder tabs was used. The ThinPot variant was used as its width is a perfect fit for the narrow strip of plastic above the Speak & Maths keyboard. The cable routing is easier if it passes through the case on the left hand side just beyond the centre line of the speaker. The ThinPot is used upside down in this placement. The wider SoftPot could be used but would have to be placed on the lower part of the case with a case entry point that avoids the battery compartment.

The three resistors were soldered directly to the solder tab terminals of the ThinPot and solid core wires were used to connect these to the CD2801 chip. Some (clear) heatshrink was used to prevent any shorting between these three connections. The connectivity is listed below, pin 1 on the ThinPot is marked with a white triangle, not to be confused with the black triangle near pin 3 which is the Spectra Symbol logo.

1. ThinPot pin 1 - 4.7k resistor - red wire - CD2801 pin 14 (VSS).
2. ThinPot pin 2 - 20k resistor - yellow wire - CD2801 pin 7 (RCOSC).
3. ThinPot pin 3 - 1k resistor - black wire - CD2801 pin 9 (VDD).

This order together with orientation of the ThinPot gives higher pitch on the right end and lower pitch on the left end matching the order of piano keys.

A few drill holes were made in a column by initially drilling down 1mm and then drilling at approximately 45 degree angle to reduce the degree of bending on the tail of the ThinPot. This was then filed into a small rectangle and the case was further filed a little to create a gradual curved path for the ThinPot to bend as it entered the case. This reduces the force from the ThinPot's bent plastic which otherwise might lift the left end of the adhered touch area. A ThinPot is shown in a photograph above being bent on a breadboard in a similar manner to how it will be bent for the Speak & Maths.

The ThinPot's tail was trimmed diagonally by about 1mm on each side with sharp scissors on each side to reduce the size as it entered the hole in the case. The area of the case that was going to be under the tail and the adhesive pad of the ThinPot was carefully and lightly scored with a craft knife to try to improve the adhesion. A tiny dab of epoxy resin was placed on the case under the tail and the tape was removed from the adhesive pad on the ThinPot and it was pressed into place. The hole for the ThinPot to enter the case ended up being wider than was needed. This was filled with epoxy resin with some masking tape inside the case to stop it dripping through as it set.

After the modification the CPUCLK frequency was checked and it had changed from 318kHz to 300kHz. This is equivalent to one semitone lower for the pitch of the voice. The wiper (collector) to other terminal capacitance is about 60pF for a 100mm ThinPot which is enough to affect the frequency and the likely explanation for this.

The video above shows the different sounds from the start of each game on the Speak & Maths with some pitch bending fun and one game of SOLVE IT on level 1.

If this article has inspired you to repair, modify or enhance a Speak & Spell or Speak & Math(s) here are some useful links to learn more about this.

Speak & Spell repairs and technical information:

1. fastmatt (YouTube)
2. Troubleshooting Speak & Spell - won’t turn on
3. Repair a glitchy Texas Instruments Speak & Spell Speak & Read Speak & Math - capacitor issue
4. Furrtek: Speak & Spell (french version) - reverse engineering and creation of a new ROM.
5. Ingmar's Retroblog: Speech output in the 80s – Speak and Spell (german version)
6. Datamath Calculator Museum:
7. Texas Instruments Speak & Spell
8. Texas Instruments Speak & Spell (British Voice Type 3)
9. TMC0271 - the 40 pin 4 bit (PMOS) microcontroller with VFD driver used in the Speak & Spell. The Vdd (pin 40), Vgg (pin 39) and Vss (pin 38) on the Speak and Spell are -3.5V, -12V, VBAT in the terminology used in this article and by Furrtek.
10. Instructables: Speak and Spell Power Board Repair - replacing transistors on the power supply board and working around a high resistance path in the membrane keyboard preventing the ON key turning the device on.
11. Instructables: Speak and Maths Audio Repair and Clean - speaker replacement, dealing with a lot of dirt, removing the membrane keyboard and adding a physical workaround to the membrane keyboard.
12. Retro Tech Repair: TI Speak and Maths Repair (YouTube) - track repair on the ribbon cable of the membrane keyboard, also shows opening up the membrane keyboard.
13. TI Portable Speech Lab by Stuart Conner
14. Plogue R&D blog: chipspeech Diary, Part 1 and chipspeech Diary, Part 2

Circuit Bending:

1. Hackday: Intro to Circuit Bending
2. Circuit Bending Wiki
3. Speak & Spell
4. Circuit-Bending: Build Your Own Alien Instruments (book) by Reed Ghazala, the godfather of circuit bending.
5. Casper Electronics Artistic Circuit Bending Schematic via MatrixSynth
6. THIS MUSEUM IS (NOT) OBSOLETE: Speak And Spell Breakout 1.2 A Look inside A Modular Circuit Bent Machine (YouTube) from This Museum Is (Not) Obsolete in Ramsgate, UK not too far from the Turner Contemporary Museum in Margate.
7. LOOK MUM NO COMPUTER: Making A Whacky Sound Generator From A Speak And Spell From Start To Finish - Circuit Bent (YouTube)
8. fastmatt (YouTube)
9. Fastmatt - How to Circuit Bend a Speak & Spell (Howto Show)
10. Video walkthrough instructions -Fastmatt Circuit Bent Speak & Spell with CV Pitch and Gate
11. Circuit Benders Forum
12. Davis Arts Center presents Speak & Spell: The Art of Repurposing, Circuit Bending and the Fashionably Retro - 2019 exhibition covered by Good Day Sacramento (YouTube)

More retro speech synthesis:

1. Hackday.io: Finding the hidden Speak & Spell in a Renault 25 - speech synthesis in a 1980s French car.
2. Adafruit Learn: Bringing Back THE VOICE of Speak & Spell

The festive photograph at the top of this page features the Songs for Ukraine Chorus at The King’s annual Christmas Day Broadcast 2025 in Westminster Abbey where they performed Carol of the Bells which is based on Shchedryk by the Ukrainian composer Mykola Leontovych.