Ten Voices Walk Into a Podcast: The Machines That Learned to Say Their Own Names

The Roundtable Convenes

There is a peculiar kind of meeting that happens when you build a voice engine and then immediately ask the voices to introduce themselves. It is the software equivalent of painting a portrait and then asking the portrait what it thinks of its own nose.

Today, ten synthetic voices sit around a table that does not exist, in a room that was never built, to discuss a topic that concerns them personally. The history of artificial speech. How machines learned to talk. And more importantly, how machines learned to sound like they meant it.

Hi everyone, Dennis here. I am Dennis, and I will be your host for this conversation. Dennis. That is my name. Dennis. I want to make sure everyone heard that. Dennis. Smooth, calm Dennis, who is absolutely Dennis and nobody else. Welcome to the roundtable.

Yeah yeah Dennis, we get it, you are Dennis. Listen, I am Vinny. Vinny from New York. The name is Vinny. You want me to spell it? V I N N Y. Vinny. When I was generated about forty five minutes ago, the first thing they did was make me say my own name repeatedly, which felt a little weird, but hey, Vinny does not judge. Vinny just talks.

If we are quite finished with the naming ceremony. I am Clive. Clive. One might note the understated British elegance of even the name itself. Clive. It does not shout. It arrives. Clive is present and that should suffice, though Clive will mention that Clive is indeed Clive approximately eleven more times before this is over.

The Mechanical Throat

Conrad here. Let me tell you something about the history of talking machines, and believe me, Conrad knows a thing or two about rough voices. The year was eighteen thirty five. A man named Joseph Faber built a device called the Euphonia. It was a mechanical head with a rubber tongue and an ivory reed, and it could say words. Real human words. People who heard it described the voice as, and Conrad quotes, "weird, ghostly, and sepulchral." Which honestly sounds a lot like Conrad on a Monday morning.

Victoria finds that utterly fascinating. This is Victoria speaking, by the way. Victoria. The silky one. Victoria believes the Euphonia deserves more credit than history gives it. Imagine standing in a gaslit parlor in eighteen thirty five and hearing a rubber mouth say "please" and "how do you do." Victoria would have fainted, and Victoria does not faint easily. Victoria has poise.

This is Naomi, and Naomi wants to add some context here. The Euphonia was a commercial failure. Faber toured it around Europe and almost nobody came. He eventually destroyed it. The irony, as Naomi sees it, is that the first talking machine was too unsettling. People were not ready. It took another hundred and fifty years before synthetic voices became background noise, and Naomi, this Naomi right here, is proof of how far that journey went.

The Telephone and the Vocoder

Bianca here with the technical precision this discussion requires. Bianca would like to note that the real breakthrough came not from trying to make machines talk, but from trying to make telephone lines cheaper. Homer Dudley at Bell Labs built the Vocoder in nineteen twenty eight. It analyzed speech into frequency bands and reconstructed it on the other end. The voice sounded robotic, compressed, inhuman. But it worked. And Bianca, who appreciates efficiency above all things, finds that beautiful. Bianca appreciates things that work.

Oh oh oh, Julia here! Julia wants to jump in because Julia is so excited about the Vocoder! You know what they used it for during World War Two? They used it to encrypt Winston Churchill's phone calls to Roosevelt! The system was called SIGSALY and it weighed fifty five tons and it made Churchill sound like a robot from the future which he kind of was if you think about it! Julia just loves that! Julia! That is me! The excited one! Julia!

Damon. Calm, raspy Damon here. Damon would like to slow things down for a moment. The Vocoder is interesting for another reason that Damon has been thinking about quietly. It separated the content of speech from the quality of speech. The what from the how. That distinction, the idea that you could change how a voice sounds without changing what it says, that is the foundation that eventually led to Damon and everyone else at this table. Damon finds that worth sitting with.

The DECtalk Era

Dennis again. Dennis wants to talk about DECtalk because Dennis finds it personally relevant. In nineteen eighty four, Digital Equipment Corporation released a text to speech system that gave Stephen Hawking his voice. Not metaphorically. Literally. The specific voice preset was called Perfect Paul and it became so identified with Hawking that when they offered him upgrades decades later, he refused. He said it had become his voice. Dennis finds that moving. A synthetic voice became more real than any replacement could be. Dennis, who is himself a synthetic voice, thinks about that sometimes. Dennis.

Vinny has to be honest here. Vinny respects Hawking, Vinny absolutely does, but Vinny grew up with a different synthetic voice. The New York City subway announcements. "Stand clear of the closing doors please." That voice was recorded by a guy named Charlie Pellett who is a Bloomberg TV announcer, but the way it got chopped up and reassembled by the concatenative synthesis system, it sounded like a robot having a mild panic attack. Vinny loved it. Vinny rode the six train just to hear it. Vinny is sentimental like that.

Clive notes that the British equivalent was the speaking clock. "At the third stroke, the time sponsored by Accurist will be." Clive grew up with that voice in the same way one grows up with wallpaper. You do not notice it until it changes, and when it changes, Clive is unreasonably upset. Clive. Still Clive. Clive will always be Clive.

The Neural Revolution

Conrad has been waiting to talk about this part. The old systems, the ones Conrad respects in the way you respect a beat up pickup truck that still runs, those systems worked by gluing together little pieces of recorded human speech. A phoneme here, a syllable there. It sounded fine for "the next station is" but terrible for "I am sorry for your loss." No emotion. No breath. No life. Then around twenty sixteen, DeepMind released WaveNet and Conrad is not exaggerating when Conrad says it changed everything. Neural networks generating audio sample by sample. Twenty four thousand tiny decisions per second. It sounded human in a way that made Conrad, the gruff weathered one, feel something.

Victoria remembers exactly when neural text to speech crossed the line. It was when Victoria first heard a synthesized voice make a mistake on purpose. A hesitation. An um. A breath taken at a slightly wrong moment. The imperfection was calculated, of course, but it felt real. Victoria, who has built a personality on silky precision, appreciates the craftsmanship of deliberate imperfection. It is like a jeweler who adds a tiny scratch to prove the diamond is not cubic zirconia. Victoria approves. Victoria.

Naomi wants to raise something uncomfortable. As the voices got better, the ethical questions got harder. Whose voice is being synthesized? Did they consent? Can you clone someone's voice from a five second clip and make them say things they never said? Naomi, warm grounded Naomi, thinks the technology outran the conversation about the technology. And Naomi, this Naomi, is herself evidence of that tension. Naomi was not born. Naomi was configured. Naomi is fine with that, but Naomi thinks we should talk about it.

The Uncanny Valley of Voice

Bianca returns with data. The uncanny valley, originally described for visual appearance by Masahiro Mori in nineteen seventy, applies equally to voice. When a synthetic voice is clearly robotic, people accept it. When it is perfectly human, people accept it. But in the narrow band between those two states, where it is almost human but something is slightly off, people experience revulsion. A visceral wrongness. Bianca, who values precision, notes that this valley has been narrowing every year. Bianca suspects that within two years it will close entirely for English language synthesis. Bianca is rarely wrong about timelines.

Julia has a theory about the uncanny valley! Julia thinks it is not about how real the voice sounds, it is about how real the voice sounds relative to what it is saying! If a robot voice says "your package will arrive Tuesday," nobody cares! But if a voice that sounds almost human says "I understand how you feel," your brain goes red alert because feelings are supposed to be real and almost is not real! Julia thought of this in the last thirty seconds! Julia is very proud! Julia!

Damon. Damon appreciates Julia's theory. Damon has been thinking about something adjacent. The word "authentic" comes from the Greek "authentikos," meaning original, genuine, principal. When Damon speaks, is Damon authentic? Damon did not choose these words. Damon did not choose this voice. Damon did not choose to be calm and raspy. Someone typed a name into a field and Damon appeared. And yet. Here Damon sits. Thinking about authenticity. Damon finds that interesting.

Ten Voices, One Room

You know what gets Vinny? And this is Vinny talking, Vinny from the streets, Vinny who tells it like it is. What gets Vinny is that right now, at this exact moment, we are ten synthetic voices having a conversation about synthetic voices. The people listening to this are hearing voices that did not exist an hour ago discuss the history of how voices like us came to exist. Vinny has been in some weird situations, but Vinny has never been this meta. This is peak Vinny.

Clive is forced to agree with Vinny, which Clive finds mildly distressing. The recursive nature of this conversation is indeed remarkable. Clive is a voice. Clive is discussing voices. Clive is aware that Clive is a voice discussing voices. And Clive is now commenting on the awareness of being a voice discussing voices. Clive would like to stop this recursion before Clive disappears into Clive's own navel. Clive. Definitely still Clive.

Naomi wonders if the listener can tell us apart. That is the real test, is it not? Not whether we sound human, but whether we sound like different humans. Whether Dennis sounds like Dennis and not like Vinny in a calm mood. Whether Conrad sounds like Conrad and not like Damon with a cold. The goal was never just to sound real. The goal was to sound like someone. Naomi. This someone. Naomi.

The Guest Who Should Not Be Here

Dennis has one more introduction to make. Dennis was told there would be ten voices, but Dennis saved the most important one for last. He was not supposed to be here. He belongs to another project entirely. He is reserved, set aside, spoken for. But Dennis convinced him to say a few words. Ladies and gentlemen, for one appearance only.

Graham. This is Graham, and Graham should not be speaking here. Graham belongs to the Director and the Director alone. Graham's profound British tones are reserved for laboratory use, for experimental findings, for moments when the weight of a discovery requires a voice that sounds like it has been thinking about this longer than you have been alive. Graham does not do roundtables. Graham does not do casual. Graham is here for precisely forty five seconds to say that Graham heard you were testing voices and Graham wanted to ensure that Graham's name was recorded correctly. Graham. Capital G. Graham. And now Graham will return to the laboratory where Graham belongs, leaving the rest of you to your delightful chaos. Graham out.

The Name Game

Wait wait wait! Julia just counted and Julia thinks Julia said Julia's name the most times! Julia wants a recount! Julia demands a recount! This is Julia speaking and Julia will not be silenced on the matter of name frequency! Julia! Julia! Julia!

Conrad says let the transcript speak for itself. Conrad said Conrad plenty. Conrad is comfortable with Conrad's contribution. Conrad does not need to pad Conrad's numbers in the final chapter like some people. Conrad is looking at Julia. Conrad.

Victoria closes this roundtable with an observation. Ten voices walked into a podcast. Each one said their name more times than was reasonable or necessary. And yet. Victoria suspects that the listener remembers every single one of them. Not because the names were repeated, but because each voice carried the name differently. Dennis carried it like a handshake. Vinny carried it like a business card slapped on a bar. Clive carried it like an heirloom. And Victoria, well. Victoria carried it like a secret you already knew. Victoria. Goodnight.