Soul in a Machine

Giving a robot a voice is one of the more fun problems in science fiction. For a long time the answer was simple: make it sound human, then add a few glitches so the audience knows it’s a machine. But what if a machine could get its point across without talking at all? In 1972, one director proved it could be done, and the idea went on to shape one of the biggest franchises in movie history.

These Are Not the Drones You’re Looking For…

Before Star Wars, there was Silent Running. Douglas Trumbull, who handled effects on 2001: A Space Odyssey, directed in 1972. The plot involved a lone botanist who kept the last of Earth’s forests alive on a freighter out near Saturn, with only three small service robots for company: Huey, Dewey, and Louie. They never said a word. They rolled around, tended the plants, and learned to play poker. By the end, one of them waters a tree alone in deep space which brought tears to an audience over a robot that never had a line. That was the trick. Trumbull showed that you could build a real character out of a machine with no face and no voice.

To make the drones move like something alive but clearly not human, Trumbull cast bilateral amputee actors who worked from inside shells and walked on their hands.¹ The idea traced back to Johnny Eck, a performer from Tod Browning’s Freaks (1932). The drones made their impact on the industry. George Lucas admitted they were the seed for R2-D2, and Universal even sued Fox over it, claiming R2-D2 ripped off the design. A judge threw the case out, but the point had been made: non-verbal didn’t mean lifeless. You just had to work harder for it.

The Soul in R2-D2

When it came time to give R2-D2 a voice, sound designer Ben Burtt had the hardest job in the crew. He tried recording babies at first, but babies don’t take direction. So he did it himself, cooing baby-talk into a microphone and shaping it with an ARP 2600 synthesizer. He’d write out in plain English what R2-D2 was “saying” in a scene, then perform the line while working the synthesizer’s knobs to match. The result was half human, half machine.²

Burtt has said the reason it worked so well was that a real person was buried in there. This gave the robot the sense of being alive, like he had “some sort of soul.”² R2-D2 never says a word we can spell, but we can understand him almost completely.

Playing Notes: Play short sporadic notes across octaves for interesting R2-D2-like dynamics. Increase ADSR attack for more emotion.

Non-Verbal Machines

Once Trumbull and Burtt showed it worked on screen, non-verbal robots stopped being a limitation and became a style. It caught on fastest in video games, where the hardware couldn’t spare the memory for real speech anyway. One of the earliest examples is Q*bert (1982). Q*bert’s angry gibberish wasn’t recorded. It was a Votrax SC-01 speech chip which was fed random input and spat out phonemes that never added up to actual words.³ But when he cursed? Every kid in the arcade knew exactly how he felt.

Doc Louis pulls the same trick in Mike Tyson’s Punch-Out!! (1987). The game started as a 1984 arcade smash before Nintendo brought it home to the NES. Between rounds, while Doc pats Mac on the shoulder giving his advice, you hear him “talk”—except there are no words, just a run of random, stepped pitches standing in for a voice. It’s the R2-D2 idea shrunk down to a corner-man pep talk. That stepped-random sound is a sample-and-hold effect, the same modulation trick that lives on the ARP 2600 Burtt used for R2-D2.

Playing Notes: Hold long notes in low then high registers to simulate the back-and-forth ringside banter of Mike Tyson’s Punch Out!!

From there it spread. When cartridge space and sound chips couldn’t handle real dialogue, developers leaned into beep speech and made-up languages, clipped chatter that lets a character emote without a voice actor. It’s a direct descendant of R2-D2’s beeps: a limitation dressed up as personality.

Manipulated Voice

But before machines could build speech from scratch, the trick was to take a real human voice and bend it. And the first great example isn’t even a robot. It’s the Wizard in The Wizard of Oz (1939): a booming voice from a giant floating head with nothing but an ordinary man at the controls, hiding behind a curtain. That’s the manipulated voice in a nutshell. A person made to sound bigger and stranger than they are.

The Throat Trick: Sonovox

The Sonovox, patented by Gilbert Wright in 1939, was one of the odder tools of the era. Small speakers pressed against a performer’s throat pushed sound up through the vocal cords. The performer then mouthed the words silently and their mouth shaped that sound into speech.⁴ The result was a train or a piano that seemed to talk. Disney used it for Casey Junior, the circus train in Dumbo (1941), and it starred as the talking piano on the kids’ record Sparky’s Magic Piano (1947). Because it shaped a real voice through a real throat, it sounded warmer than the vocoder that came later.

Bending the Voice Electronically

By the 1960s manipulation went electronic. The Daleks on Doctor Who (1963) got their harsh screech from a ring modulator at the BBC Radiophonic Workshop. Brian Hodgson ran the actor’s live voice through it, tuning the carrier frequency to a 30 Hz tone.⁵ A real performance chopped into something inhuman. The same circuit built the electronic score for Forbidden Planet back in 1956.⁶

The vocoder took over from there. It splits a voice into frequency bands and uses them to shape a bank of oscillators. You can still make out the words, but they don’t come from any living being. One of the earliest examples is Colossus: The Forbin Project (1970), where a supercomputer builds a voice for itself. Behind the scenes, that was actor Paul Frees, run through processing until he sounded cold and machine-made.⁷ The most famous version showed up in Battlestar Galactica (1978). The Cylon voice came from an EMS vocoder paired with an ARP 2500 synth.⁸ An actor read every line in a flat monotone so he wouldn’t throw off the vocoder. The rig was so touchy that a power flicker could change the sound overnight, and there’s only about fourteen minutes of Cylon dialogue in the whole original series. The same tool built Soundwave, the Decepticon in The Transformers (1984), who stood out from the cast because his voice was so obviously processed.⁹

Playing Notes: A more experimental patch, triggered by either keyboard or voice, or both. Keep in mind that note on events will process audio input, irregardless of what’s connected. If you’re not prepared this could lead to some feedback, though maybe that’s just what you were looking for.

Synthesized Speech

So far we’ve looked at bending real voices to sound like machines or filling in for a voice with beep speech. But the Votrax chip in Q*bert was a speech synthesizer at heart, fed nonsense on purpose. The last step, logically, was creating a machine that builds real words from nothing—no actor, no recording. Two approaches won out, and the difference between them came down to soul.

The Machine That Reads

The first approach and purest version was rule-based. No stored voice, no recorded phonemes. Just a big set of rules that models how a human throat, mouth, and nose shape sound. Feed it any typed text and it reads forever. This was the life’s work of MIT’s Dennis Klatt, whose research became the DECtalk hardware.¹⁰ You may have heard it before, it’s the voice of Stephen Hawking. The default DECtalk voice, “Perfect Paul,” was modeled on Klatt’s own voice.¹⁰ Even the coldest machine voice had a real person inside it. It’s the perfect voice for a robot with no soul, which is exactly why this approach later got cast as one.

Talking Silicon

The other approach was to squeeze a small vocabulary onto cheap ICs. In 1980, compressing computer speech cost about a thousand dollars per word, which is why the robots in Berzerk (1980) had a thirty-word vocabulary they reused over and over: “Intruder alert! The humanoid must not escape!”¹¹ Gorf and Wizard of Wor (1981) used the phoneme-based Votrax SC-01. And in 1985, Gauntlet‘s “Warrior needs food badly” ran on the Texas Instruments TMS5220C, the same family of IC that taught the Speak & Spell how to talk.¹²

An honorable mention, while Sinistar (1983) sounds synthesized (“Beware, I live!”), it isn’t. That’s a real voice that radio announcer John Doremus recorded and compressed through the HC-55516 and then dropped in pitch to sound like a monster.¹³

Where It Comes Together

WALL-E (2008) put Ben Burtt back on the same problem he cracked with R2-D2, thirty years later. Using a synth program called Kyma, he again started with his own voice, broke it apart, and rebuilt it until “WALL-E” and “Eeee-vah” sounded like two machines in love.¹⁴ His goal, in his words, was to “create the sense of a soul” with sound.¹⁴ Same method, same words, as 1977. And here’s the interesting part. The villain, AUTO, the ship’s autopilot, was voiced by MacinTalk—Apple’s built-in text-to-speech tool, a cousin of DECtalk from that same 1984 wave of reading machines. Director Andrew Stanton picked it on purpose. He wanted AUTO to sound “cold, zeros and ones… and soulless.”¹⁵

It’s the evolution of the robotic voice right there in one movie. A machine can be made to talk a lot of ways, it only comes alive when there’s a person hiding somewhere in the sound.

---

Notes

1. “1971 – Silent Running Drones – Doug Trumbull, Don Trumbull, Paul Kraus, James Dow (American).” Cybernetic Zoo, https://cyberneticzoo.com/not-quite-robots/1971-silent-running-drones-doug-trumbull-don-trumbull-paul-kraus-james-dow-american/. Accessed 16 July 2026.
2. Ben Burtt on building R2-D2’s voice and giving it “some sort of soul.” “Ben Burtt: The Sound of Revolution and Creating Icons.” Fantha Tracks, https://www.fanthatracks.com/news/film-music-tv/ben-burtt-the-sound-of-revolution-and-creating-icons/. Accessed 16 July 2026. “Ben Burtt.” Wookieepedia, https://starwars.fandom.com/wiki/Ben_Burtt. Accessed 16 July 2026.
3. The Votrax SC-01 phoneme chip in Q*bert, Gorf, and Wizard of Wor. “Noise Hits All at Once.” Flow, Dec. 2025, https://www.flowjournal.org/2025/12/noise-hits-all-at-once/. Accessed 16 July 2026. “Votrax SC-01.” Vocal Synthesis Wiki, https://vocal-synthesis.fandom.com/wiki/Votrax_SC-01. Accessed 16 July 2026.
4. The Sonovox (Gilbert Wright, 1939; Casey Junior in Dumbo; Sparky’s Magic Piano). “Talk Box.” Wikipedia, Wikimedia Foundation, https://en.wikipedia.org/wiki/Talk_box. Accessed 16 July 2026.
5. Dalek voices via ring modulator at the BBC Radiophonic Workshop. “The Story of the BBC Radiophonic Workshop.” Sound on Sound, https://www.soundonsound.com/people/story-bbc-radiophonic-workshop/. Accessed 16 July 2026.
6. Ring modulation in the Forbidden Planet score. “Ring Modulation.” Wikipedia, Wikimedia Foundation, https://en.wikipedia.org/wiki/Ring_modulation. Accessed 16 July 2026.
7. Colossus voiced by Paul Frees through electronic processing. “Colossus: The Forbin Project.” Wikipedia, Wikimedia Foundation, https://en.wikipedia.org/wiki/Colossus:_The_Forbin_Project. Accessed 16 July 2026. “Secret History of the Vocoder.” Gearspace, https://gearspace.com/board/electronic-music-instruments-and-electronic-music-production/1141251-secret-history-vocoder.html. Accessed 16 July 2026.
8. The Cylon voice (EMS vocoder + ARP 2500). “Robot Voice Design: How Joe Grandberg used analog gear to recreate the Cylon voice from Battlestar Galactica.” A Sound Effect, https://www.asoundeffect.com/robot-voice-battlestar-galactica/. Accessed 16 July 2026.
9. Soundwave’s vocoded voice. “Battlestar Galactica And Transformers Are Linked By One Unique Special Effect” SlashFilm, https://www.slashfilm.com/2120729/battlestar-galactica-transformers-link-vocoder-special-effect/. Accessed 16 July 2026.
10. Dennis Klatt, DECtalk, “Perfect Paul” (based on Klatt’s own voice), and Stephen Hawking. “Stephen Hawking’s famous voice belonged to this pioneering MIT scientist” Big Think, https://bigthink.com/high-culture/this-mit-scientist-gave-his-voice-to-stephen-hawking/. Accessed 16 July 2026.
11. Berzerk‘s LPC speech and thirty-word vocabulary. “Berzerk.” Arcade History, https://www.arcade-history.com/?page=detail&id=236. Accessed 16 July 2026. “Berzerk.” Video Game History Wiki, https://videogamehistory.fandom.com/wiki/Berzerk. Accessed 16 July 2026.
12. Gauntlet‘s TMS5220C narrator (1985). “1985: 10 Things You Might Not Know About Gauntlet.” Warped Factor, June 2015, http://www.warpedfactor.com/2015/06/1985-10-things-you-might-not-know-about.html. Accessed 16 July 2026.
13. Sinistar‘s digitized voice (John Doremus, HC-55516 CVSD). “Sinistar.” Wikipedia, Wikimedia Foundation, https://en.wikipedia.org/wiki/Sinistar. Accessed 16 July 2026.
14. WALL-E‘s Kyma-processed voice and Burtt’s “sense of a soul.” “Thinking through sound — Ben Burtt and the voice of WALL-E.” Symbolic Sound, Sept. 2024, https://news.symbolicsound.com/2024/09/thinking-through-sound-ben-burtt-and-the-voice-of-wall-e/. Accessed 16 July 2026. “How Did They Make That Sound on WALL-E?” Animation World Network, https://www.awn.com/news/how-did-they-make-sound-wall-e. Accessed 16 July 2026.
15. AUTO voiced by MacinTalk. “WALL-E.” Wikipedia, Wikimedia Foundation, https://en.wikipedia.org/wiki/WALL-E. Accessed 16 July 2026.