REID HOFFMAN: Let me test the zoom levels. Testing one, two, three. Okay, those seem to be all right. 

JEFF BERMAN: That’s your founding host of Masters of Scale, Reid Hoffman. He’s at his home office, with a laptop open and a microphone setup. Two producers are standing by in their zoom squares. 

HOFFMAN: Let me get the coffee and water. Actually, I guess coffee is not always great on the voice. Well, I’ll still have a little bit of it. 

BERMAN: He settles in his chair, ready to record some voiceover for the show. 

HOFFMAN: Just making sure that it works. Here we go. And then this out of the way. Quick time. Do, do, and two little pickups. 

BERMAN: Reid turns to the script and starts recording. Often, it’s smooth sailing. But sometimes… 

HOFFMAN: That’s Jesse Berry, an ornithologist at the Cornel Blah. That’s Jesse Berry, an ornithologist at the Cornell Lab. Blah, blah, let me just make sure…

That’s Jesse Berry, an ornithologist at the Cornell Cornell Lab. 

BERMAN: Reid is an investor, an entrepreneur, a philosopher, a patriot, and a great friend. But he’d be the first to say: he is NOT a voiceover artist.

HOFFMAN: Over time, the defini. Over time, the definition. Over time, the definition expanded to refer to anyone, real or imaginary…

BERMAN: Recording voiceover can be time- and labor-intensive, even for the pros. At Masters of Scale, we might update scripts at any time, as stories develop. So when Reid travels — which is often! — he takes a microphone with him. And he needs a super quiet space wherever he is. 

Good voiceover work also means getting into the right mindset. When you hit record, you have to match the tone of the story at hand, no matter what’s going on in your life. Not to mention everyday hazards to recording — like a frog in your throat, a dog barking outside, or suffering from a cold that has you sniffling between sentences.

But imagine you could ensure a result like this every time. No stumbles, no X factors.

REID-ISH: We rightly value learning from different perspectives. Doing so gives us insight and wisdom we can then incorporate into our own stories. 

BERMAN: That is an AI version of Reid Hoffman’s voice. About a year ago, Reid dove into fully cloning his voice through a process powered by artificial intelligence. And, like a lot of AI technologies, it’s improved dramatically in a short amount of time. It’s basically indistinguishable from human Reid sitting in real time at the microphone — even to Reid himself.

REID-ISH: I’m Reid Hoffman, co-founder of Linked-in, partner at Greylock, and your host.

BERMAN: So, we want to tell you/share the story of this technology — to take you under the hood, to better understand it. You’ll hear from Reid and me, from our production staff, and from the voice cloning company we’re working with, called Respeecher. Voice cloning is a tool that offers scaling potential for so many things, including this very show. It’s also a tool with major implications we must grapple with in audio and beyond.

I’m Jeff Berman, your host. 

One of the goals here at Masters of Scale has always been to democratize lessons in entrepreneurship, rooted in the experience of LinkedIn co-founder and legendary investor, Reid Hoffman. We do in-depth interviews with business leaders; the scheduling of and preparation for those interviews can take months. After the interview comes an intensive process of editing, writing, sound design, and recording scripts of voiceover to connect the dots of the episode. 

The voiceover step can be tough for a lot of reasons. But now, we have a full clone of Reid’s voice. And we are exploring how to use it on Masters of Scale. 

This is a synthetic voice, trained on the sound and cadence of Reid speaking.

REID-ISH: Hi listeners, If you’re interested in experimenting with AI like we are, then I want to invite you and your team to join us this week on February 29 for Masters of AI day.

BERMAN: It replicates his voice, but it doesn’t generate original content.

As an early adopter of new technology, Reid not only cloned his voice. He also made a generative video version of himself, called Reid AI.

REID AI: Hello everyone, I’m thrilled to be here today. I’m an AI-generated version of Reid Hoffman, his digital twin. My thinking and everything I am saying comes from a custom GPT that is based on 20 years of Reid talking and making content. So I’ve read all his books, watched his speeches, and listened to his podcasts.

I think it gives me a pretty good sense of how you approach the world, Reid.

BERMAN: In recent videos, Reid appears on a split screen, going back and forth on camera with an expressive, vivid digital twin. This generative AI has been trained on all the content Reid has produced over the years — like books, speeches, and podcasts. The personality of Reid AI comes from a custom chatbot based on all those hours and pages of material, entered into Chat GPT-4.

HOFFMAN: So first, I’m going to challenge Reid AI with some questions to see what it can do. Let’s take a real piece of work from me. Blitzscaling. I wrote a 336-page book, and how do you explain it to a five-year-old?

REID AI: For a five-year-old, blitzscaling is like building the biggest tower super fast before anyone else, even if it gets a bit messy, so you can be the leader of the tower game.

Reid Hoffman on creating his AI twin

BERMAN: As a listener, you might already be trying to figure out in your heart which Reid is which. Is what you’re hearing one of several synthetic versions? Or is it Reid, live and in-person? Well, I sat down with Reid — human Reid, in real time — to talk more about why he’s exploring all this so ambitiously.

HOFFMAN: Ready when you are.

BERMAN: Hi Reid!

HOFFMAN: Hi Jeff.

BERMAN: We are getting together just days after you released kind of a mind-blowing video of you interviewing an AI version of yourself and that AI version of yourself interviewing you. I’m curious how the idea for that came about.

HOFFMAN: So the idea was, obviously most of the time that people have been talking about digital twins, most of that discussion has been in fear and, “oh my God, it’s going to be the end of Hollywood,” and, “all this disaster and storm is coming.” So the idea was to start exploring what some of the positive use cases are. Sharing the journey, sharing the learning, sharing the exploration, With an emphasis on what can be possible and what can be positive. But of course navigating concerns and experiences as well. 

BERMAN: Because, you know, for as long as you’ve been hosting Masters of Scale, I know it’s been a labor of love. But there’s a labor part of it. And so could you talk a little bit about what we’re experimenting with a synthetic Reid AI voice?

HOFFMAN: Look, if I can figure out how to have tools, make it a lot more efficient for me in terms of how I do work and all the rest, that’s a good thing.

I use a phone for that. I use computers for that.

So for example, if you say, well, look, why am I using this as well? Because I can address commentary that I otherwise don’t have time for, or I can show some lens into the future, which is helpful to our audience. Or, I can go into more depth and produce some more content that hopefully some number of people will really like that I wouldn’t have otherwise have time to do.

Because by the way, I sometimes have somebody who writes a speech for me, and I give the speech.

And no, I don’t go, and by the way, sentence X was from Susan, and sentence Y was from Bob, you know, I go, okay, I’m giving the speech. Now, of course, I give the speech, like you can then yell at me if I’ve done something wrong, and I’ve worked with a team to hopefully get it right. But like, you don’t have to go, Susan generated line three and Bob generated line seven, or GPT generated line 10.

BERMAN: So, when you think about taking a synthetic AI voice and being able to approve a script and then have it read and have it be indistinguishable for the audience, that feels like a tremendous use case here.

HOFFMAN: A thousand percent. General principle is if you’re doing it — legitimately trying to do more or valuable things for other people — then I think you will be broadly on the right path.

BERMAN: An AI voice clone allows Reid, and Masters of Scale, to fulfill our mission of getting more business lessons and storytelling out in the world.  It has trade-offs — and we will get to those. But there’s also value, including for our producers.

Masters of Scale producers on the value of cloning Reid’s voice

EVE TROEH: Delivering successful voiceover is such a remarkable talent and skill. 

BERMAN: That’s our Masters of Scale executive producer, Eve Troeh. 

TROEH: But there’s also just… The statistical probability that someone who has brilliant things to say and important lessons to share will be good at that specific skill of reading text from a page and deliberating in the voice — it’s very low.  And it can take hours to deliver a couple paragraphs. 

Oof. It’s hard work, though.

BERMAN: Also, there’s the matter of vocal quality. Here’s producer Alex Morris. 

ALEX MORRIS: I’m thinking about specific times where Reid has had a cold, and we’ve said maybe we should pause recording because even though this is naturally organically Reid right now speaking, it doesn’t sound like the Reid Masters-of-Scale-Reid voice that we know.

BERMAN: Then there are the last-minute pickups.

TROEH: We’re talking about accurate information here. 

Often to get the information exactly right, whether it’s on business, or to get the pronunciation of someone’s name, you may have to, and I have had to, wake up a host at a very odd hour, call someone, text them, at two in the morning, so that they can say three words, so that the story going out can be accurate. And is that the best use of everyone’s time and energy?

To me, if there were a magic bullet that could change just those three words so that they could be correct in the host voice, I think that would be a good use of that magic bullet. 

Respeecher founder/CEO on creating the cloning tech

BERMAN:  Well, that magic bullet seems to be here. And it’s courtesy of a company based in Ukraine, called Respeecher. They’re the ones who cloned Reid’s voice. And they do not underestimate the impact of their work:

ALEX SERDIUK: The basic limitation of a human being in front of microphone is now removed.

BERMAN: That’s Alex Serduik, founder of Respeecher and the CEO. The idea for Respeecher started at an IT hackathon in Kyiv. Most participants were focused on visual recognition. But Alex and his co-founder had a different goal.

SERDIUK: To create a very simple model that would let one person speak in the voice of another person 

BERMAN: They won the hackathon. And a few years later, their model grew into Respeecher, a voice-cloning company that creates synthetic versions of actual human voices. A lot of their projects involve TV and movies — remarkably often bringing dead voices back to life.

ELVIS CLONE: Thank you, thank you very much. It sure is an honor to be here with y’all on America’s Got Talent.

SERDIUK: We were on America Got Talent in the voice of Elvis Presley.

ELVIS CLONE: I have to tell you we are gonna do a couple of our biggest records tonight.

SERDIUK: We were at Super Bowl in 2021 in the voice of Vince Lombardi. 

VINCE LOMBARDI CLONE: Because it’s not whether we get knocked down. It’s whether we get knocked up.

BERMAN: Even actor Jimmy Stewart, whose voice has been revived for a bedtime story on the meditation app Calm.

JIMMY STEWART CLONE: Well, hello, I’m James Stewart. But well, you can call me Jimmy.

BERMAN: Respeecher works closely with the companies that own these brands and intellectual property. And if they work with the voice of someone who’s died, they cooperate with their estate. 

And there are applications beyond entertainment. Respeecher helps individuals with medical complications. 

SERDIUK: We can change quality of lives of people who have speech disorders. Back in 2022, we tried our first rough real time models on patients with laryngectomy.

BERMAN: And those who’d be in danger if their real voice were heard.

SERDIUK: You often need to anonymize victims of a crime for a documentary or whistleblowers. Usually it’s been done with voice morphers, which make it sound bad, like really bad. It’s really hard to understand the sound and emotions.

BERMAN: From his home in Kyiv, Ukraine, Alex can even imagine Respeecher helping victims of war crimes tell their stories more safely. And he’s proud of his company’s ethics. He said they’ve had no reported cases of misuse, and they don’t take clients for political ads.

Last year Reid hired the team at Respeecher to create a synthetic voice for himself. 

There are hundreds of hours of Reid hosting Masters of Scale. So data was not a problem. 

But Respeecher doesn’t turn text into speech — that’s the technology behind vocal assistants that might sound more robotic, like Alexa and Siri. Respeecher uses human voiceover artists who deliver the lines. 

SERDIUK: We understood that, look, let’s take from humans what they do best. Humans are amazing at performing because we were born with this vocal apparatus and use it on a daily basis.   

BERMAN: Voiceover artists perform with emotion, unique pauses, and other quirks. Then that audio file gets a filter, powered by the AI model, to sound like the actual person you want to replicate. Here’s an early draft of Reid’s synthetic voice.  

DRAFT OF REID-ISH: For Isaac Newton, it was an apple falling from a tree that supposedly led him to formulate the law of universal gravity. 

BERMAN: A little stuffy, right? A bit off? Here’s a further iteration from Respeecher. 

DRAFT OF REID-ISH 2: For Benjamin Franklin, it was a key on a kite string struck by lightning in a thunderstorm that proved electricity could be transferred and stored. 

BERMAN: Deeper, more casual, more like Reid. 

The original use case for this was to make an audio-book, specifically Reid’s 2023 work called Impromptu: Amplifying Humanity Through AI. So, using an AI voice to record a book about AI — makes sense, right? Listeners got the voice of Reid, without the dozens of hours it would have taken him to record reading it. 

Here’s a sample of how it turned out.

REID-ISH: My query to GPT-4 was not the first time I’d asked an LLM to create a lightbulb joke for me. Earlier GPTs can freeze up like an amateur at an open mic night.  How many restaurant inspectors does it take to change a lightbulb? GPT-3…

GPT-3: Only one, but the lightbulb has to want to change.  

REID-ISH: That robot should keep its day job.  

BERMAN: But the technology has come even further since then.

SERDIUK: When we did the book, we used models we most probably don’t use anymore. We either improve the quality of the sound or the update is related to speed or usability or time to train those types of things.

When we have to listen to a sound we created a couple of years ago, we dislike it. We really feel bad, we want to remake the voice, and that’s the case when we could reach out to a production company and just say, can we please do it again?

Just replace the vocal track, we would do it for free, we just want it to be better on the current level of quality.

BERMAN: Coming up, the Masters of Scale team does a test drive with the latest version of Respeecher. 

REID-ISH: And so, how British do you want me? More than normal?

BERMAN: And we ponder the vast implications. That’s after a break.

[AD BREAK]

Creating Reid-ish

BERMAN: Ok before we get even deeper into the world of Reid’s vocal clone, it only seems fair to give it a name.

Reid, I mean you’ve referenced your digital twin. Is that how we should think about an AI-generated Reid voice or, or would you want us to think about it differently? 

HOFFMAN: Digital twin’s okay. That wasn’t really the language I generated. That was the team doing it. It’s kind of a little bit of an industry parlance that I think is a little strange. It is good to have something that is both like Reid but also Reid-ish. ‘Cause there’s English in a Reid-ish. Ish is kind of a Suffix that people say, do you think that this is a complete catastrophe?

It’s like, well, ish.

BERMAN: Love it. Reid-ish, it is.

Now, to better ponder what we’re truly getting into, producers Eve Troeh, Juliette Luini, and Alex Morris logged into our Respeecher account and got to work.

TROEH: Okay, I am rolling again. So we’re making these recordings, and then we’re going to feed them into Respeecher. Okay. 

It’s kind of nerve-wracking. 

JULIETTE LUINI: I know. I’m sweating like Reid probably does in his recording sessions. 

TROEH: Alex, are you ready to play Reid Hoffman?

MORRIS: I’m ready. 

BERMAN: Our producers tried out the speech-to-speech process, hearing how realistic Reid-ish sounds filtered through each of their voices. 

TROEH: “I’m Reid Hoffman” is a strange thing to say because I’m not Reid Hoffman. But, this is the technology we’re thinking of using. So, I’m going to go for it. 

TROEH: I’m Reid Hoffman.

LUINI: Co-founder of LinkedIn. 

MORRIS: Partner at Greylock. 

LUINI/MORRIS/TROEH: And your host. 

TROEH: And I believe… 

MORRIS: And I believe… 

LUINI: And I believe… 

TROEH: We’ve uploaded our audio files, all three samples, and now, tell me where we’re at. 

LUINI: Now it is converting our original audio file into Reid Hoffman. And it’ll take like five minutes. 

MORRIS: That’s kind of surprising to me, like five minutes hopefully. I genuinely thought it was going to be something where it’s like 24 hours, and then they send you some sort of a file like, I am shocked like that’s pretty close to real-time. 

TROEH: Yeah, it’s like about the amount of time you would need to upload a big audio file and convert it to any other format.

LUINI: It’s ready! Do you want to just play it?

TROEH: Let’s do it. 

LUINI: Excited to hear how it picks up Alex’s accent.

MORRIS: I’m Reid Hoffman… 

REID-ISH: … co-founder of LinkedIn, partner at Greylock, and your host. And I believe you need to amplify the unique stories of your community and your customers. Doing so will build a foundation of trust, loyalty, and enduring impact. 

MORRIS: Wow. Okay. My first thought was that hearing Reid’s voice with my specific accent, it sounded to me like he was like a London gangster. It was like a Guy Ritchie film. He was like one of the Kray Twins. I’m honestly so impressed at how, I’m literally like: that was Reid Hoffman with a British accent.

If he was making fun of me, he would do that. Obviously, for an American to do a British accent is a very safe voice that you can do, right? In terms of cultural appropriation. We’re gonna have to start having conversations about where’s the line? What can you do? And I don’t know what feels insensitive.

TROEH: Yeah, that’s a good point. It brings up these issues of authenticity. You can pretend to be another nationality, another gender, that you are not. And it’s so easy as a media producer, I’m immediately so excited about the technology that I need someone to prompt the discussion of the dangers of it. Like that question of authenticity and identity. 

Then I think, oh, right. Yes, that’s right. This could be used for other things that are not so great, but because we’re so focused on what we’re aiming to use it for, I think it just makes me pretty giddy.

Okay, let’s hear yours.

LUINI: I’m Ried Hoffman…

REID-ISH: … co-founder of LinkedIn, partner at Greylock, and your host. And I believe you need to amplify the unique voices of your community and your customers. Doing so will build a foundation of trust, loyalty, and enduring impact. 

LUINI: Okay, we are all holding back explosive laughter right now. That was unpleasant to listen to. 

TROEH: I wouldn’t say it was unpleasant to listen to. It just didn’t sound, didn’t sound like Reid. 

MORRIS: He sounded so scared. He was like, okay, well, here I go, I guess. I know.

REID-ISH: Okay, no pressure. Okay. I’m sweating like Reid does in his recording sessions. Okay, I’m Reid… that is weird to say.

LUINI: But it also did sound like my voice had changed. Undertone of fear in it, which I don’t know if that comes across day-to-day. 

TROEH: You’re rethinking everything you’re thinking. Do I sound like a fearful woodland creature in my everyday conversation? No, I don’t think you do.

MORRIS: Yeah, I mean, that’s what’s nuts, is I think I thought this was gonna feel like we’re doing an impression of Reid Hoffman, but I actually felt like Reid was doing a mean impression of us.

LUINI: Yes. That’s so accurate.

TROEH: Should we hear the other one? 

MORRIS: Eve’s!

TROEH: I’m Reid Hoffman…

REID-ISH: …co-founder of LinkedIn, partner at Greylock, and your host, and I believe you need to amplify the unique stories of your community and your customers. Doing so will build a foundation of trust, loyalty, and enduring impact. 

TROEH: Wow. I can tell that it’s me because I know how I say things and also my cadence. But there are a few places where it does make me laugh. Sounds like Reid. I don’t know, what’d you guys think?

LUINI: I want to play with the pitch and see if I can move the pitch down because I think you, Eve, did a very good Reid Hoffman actually so I clicked a plugin, and it says regeneration settings, pitch shift, and it has a scale from negative 12 all the way to positive 12, and we can play around with this scale to change the pitch. If we did like a negative six, I feel like it would be really good.

MORRIS: I feel like that’s going to sound like spot on, actually. So, maybe we should try that now?

LUINI: Yeah, let’s do it. 

TROEH: I’m Reid Hoffman.

REID-ISH: Co-founder of LinkedIn, partner at Greylock, and your host. And I believe you need to amplify the unique stories of your community and your customers. Doing so will build a foundation of trust, loyalty, and enduring impact. 

LUINI: That’s good. That was really good. I feel like you get the sort of downward ending of sentences like Reid does, and that’s where it really sounded like him.

MORRIS: That to me was pretty spot on. I think that’s the best one so far for sure. 

REID-ISH: Picture a vast library with sections for every genre you can imagine. You can find escapism in the romance and sci-fi sections. points of reflection in memoir and poetry, and guidance in the self-help and business sections. 

MORRIS: I, to be quite honest, still don’t know how I feel about it. I think we were in a really interesting position because, you know, we’re working with one of the pioneers of AI technology, it means we’re encouraged more than I think most places to kind of experiment with it. 

Even when you’re experimenting, I think it still feels like that there’s stakes there and feels like something that we wanted to make sure that we were always being fully conscious of what we were doing and what we were creating.

TROEH: One of the ways that we’re putting guardrails on the use of the technology is that the script is still written by writers, researched by researchers, vetted by Reid and signed off on editorially by him. And we’re really putting a lot of importance on that and feeling that that’s a big part of the integrity of the editorial process and the production process that gives us license to use Respeecher because we are really thinking of it as a technical production tool rather than displacing any intellectual or creative labor.

DAN NEILAN: I think that’s a great point. And I think that is a really important line to draw.

BERMAN: Staff writer Dan Nielan chimes in on that front.

NEILAN: Because generative AI that is creating content gets a lot of the headlines. When people hear AI, that’s immediately what they think of. They think of a model that is producing written word or audio or video, whatever it is. Underlining that the AI we are using is purely a production tool, and the intellectual content of the show is remaining the same, and is remaining directly tied to Reid. 

Really making sure the audience understands that is a huge thing for me in particular as a creative and as a writer, because I want people to know we’re not handing over the keys to the car. We’re just, like — I can’t extend this metaphor anymore — changing the paint job?

BERMAN: Associate Producer Masha Makautunina was also concerned about ethics.

MASHA MAKUTONINA: In the English language, there is a phrase, you’ve got my word. Maybe it was read, maybe it was approved by you, but not technically said by you. And suddenly it’s Pandora’s box. 

But my relationship with this technology also was affected when I heard the way Reid talks about it. And I really, really, really believe we need to approach this as case by case point. And when I hear the case that Reid makes for the use of his voice, it completely changes my relationship with the person whose voice we are using, being so enthusiastic about it. I think it’s the main puzzle piece that is important for this case. 

BERMAN: Reid has been releasing digital twins of himself into the world, one by one, whether it’s our Reid-ish voice from Respeecher or the full generative video presence he calls Reid AI, from Eleven Labs and Hour One. 

HOFFMAN: The value is you can do a bunch of things where I can have superpowers that I don’t. For example, I would love to speak more languages.

HOFFMAN: I am not going to spend the time in order to do it. But the ability to, say, hey, I’ve got content that might be interesting or useful, like about entrepreneurship and a number of different languages.

To be able to make that more humanized into, what is Reid like when he speaks Chinese or French or Italian or Spanish or German? All of those things are I think good things as opposed to the, hey, it’s so I can care less and pay less attention, it’s so that I can actually accomplish more in ways that matter for people. That’s, I think, the general use of everything from Reid AI down to synthetic voices. 

What do the listeners have a right to know?

BERMAN: Even for the positive use cases, the deployment of a synthetic voice clone raises important questions around disclosure – whether and when people should know they’re hearing AI-generated audio. This entire behind-the-scenes episode exists because we are committed to transparency. But those standards are hardly codified in our industry.

One of the big questions that we’re wrestling with, and I know you’re wrestling with, and hopefully everyone in this industry is wrestling with, is what? What do the listeners have a right to know? So when you think about a listener’s bill of rights, what do you think is in that? 

HOFFMAN: One principle is truth telling slash anti deception. If you are using a digital twin of your image or voice, you should be signaled in some way. It doesn’t necessarily have to be like, you know, like glaring red lights in the background, right?

But like, clearly not hidden, because we’re storytelling creatures, I don’t think it has to be like an exact accounting. These three sentences were non-digital Reid. And these three sentences were Reid himself. And it’s like, no, no, no, no, you can choose how deeply you go.

It’s kind of more of a, okay, you’re being forthright with me about how you’re producing. 

BERMAN: Is that sort of an equivalent to almost a “names have been changed to protect the innocent” disclaimer at the top of a film or something like that? Is that how you think about that?

HOFFMAN: Exactly. Or, even this is based on real events, but has been fictionalized, it’s like, okay, great. You’re making sure that I don’t go, it is all real, right? Or it is all fiction. Right? You’d have to sort it out if it matters to you, you’re gonna have to look into it some. 

It’s a question of do you stand by it?

Movies don’t go: “And there were computer graphics in this film.” It’s like, we all know there’s computer graphics. They don’t actually say this film or that film, because that’s a natural part of the production process.

If something is generated, it says, Reid says X and Reid never looked at it, then it’s like, well, we can’t stand and say, Reid said X, right? Like we say, Reid says X all the kinds of time we’ve done in a Master’s of Scale for every single episode, which has been written by a writer. 

Just to be clear. This is not new ground for us.

Look. Do I own it in that way and as an authentic offering to people and that really really matters? Are we legitimately and with high integrity, trying to be really, really good to our community and our listeners?

BERMAN: One thing is clear: we’re not underthinking this. We have this ground-breaking technology ready to go. We have signoff, even enthusiasm from the source. We are going to use it. And we are going to disclose it. Whether that’s in the episode audio, show notes, or some embedded metadata option that comes along, we don’t yet know. The conversation is just starting. We hope you’ll stay tuned.

REID-ISH: This is Reidish, signing off. Or is it? No. It is. I’ve come a really long way.