How I Actually Make AI Voice Sound Real

So, in this vid- no, wait. So, in this video, we’re going to… break down the… Ugh, that sounds terrible… Still recording?

I can’t get it right. I just don’t sound… like a youtuber.

Good.

Good? What’s good about that? Have you heard an advice, “if you’re homeless, just buy a house”. What the f? That doesn’t make any sense.

I mean if you don’t sound like a youtuber, then just use something that does.

And just like that, I went down the rabbit hole of… AI voice generators.

So… this is how I actually make an AI voice sound… real… Okay, look. Before we get into the technical stuff, if you wanna make a realistic voice with

Okay, look. Before we get into the technical stuff, if you wanna make a realistic voice with AI, you need to understand what the hell makes human voice realistic in the first place. Really,

like why does your brain think you're listening to a real person right now instead of a computer pretending to be a ginger dude with a fancy suit? Well, if you break it down, there are 4 elements.

If I keep the same tone throughout the entire video with no ups and downs in my voice, that would make it sound robotic. I talk like that in real life… Oh, man, this robot can talk more human than me? Anyways! If all of a sudden I change my tone from monotonous to super excited, that would make it sound more human.

You probably heard that cutting all the pauses in your video boosts retention. Even though

that’s kinda right, it’s not worth making yourself look like a lifeless robot that’s been controlled by the algorithms. Instead, pause on some important points… to make your message… more impactful… And make your AI voice sound more natural, of course.

Just like making pauses, emphasizing important words in your script… no… EMPHASIZING important words in your script, is also a powerful way to make your AI voice sound more realistic.

And lastly, of course, if you use ChatGPT to write your entire script, no matter how realistic your AI voice is, at some point, it will start to sound robotic. And yeah,

if you write your scripts with AI, and think that nobody knows… trust me… we know… Alright, so with these 4 elements, you can cook a 96.2% pure ai voice. These

are the main things you need to keep in mind while creating your voiceovers. And

we’ll use each of this in a minute while creating the actual voiceover.

But before that, we need to… Choose a software.

Look, there are so many AI voice generators out there right now that I just gave up counting. I

tested almost every single one for this video, and you know what? Most of them are… fine. They do a decent job at replicating a human voice. But the one I always come back to,

are… fine. They do a decent job at replicating a human voice. But the one I always come back to, is called, ElevenLabs. So that’s gonna be the software we’ll use for this video.

And no, they didn’t pay me to say this. I wish they did, but they didn’t.

I’ve been using it since I started this channel, and you’re seeing… or hearing how realistic it sounds. Be honest. You

sometimes forget it’s not a human voice, right? You… you don’t? Okay, nevermind…

There’s a link in the description if you wanna check it out and follow along with your own script. Just create and account, and try the techniques we’re gonna use by yourself. Okay, fine. Let’s do this.

So you know the theory. Now, what do you need? A voice!

This is an important step. Because once you pick a voice, you're basically married to it. All your future videos. That voice. So don't just click the first one you see and

it. All your future videos. That voice. So don't just click the first one you see and call it a day. You’re gonna need something solid. Take your time. Pick a nice one.

Now there are a few ways to create a voice.

First, if you wanna use your own voice, you can go with this route.

You may think “why would I clone my own voice when I can just… use my throat”, but it’s gonna be a really important step for the end of the video. Just let me cook.

Click the ‘voices’ tab. ‘clone a voice’, then ‘instant voice clone’, and upload a sample.

If you got time, you can create a professional voice clone as well.

But keep in mind that it requires at least 30 minutes of voice sample. Once

you generated your voice clone, “you can use it wherever you want. that’s crazy”.

Wait, actually… yeah, that’s completely illegal.

You can’s clone someone else’s voice without their permission. It’s like,

against every terms of service ever written. So you better use it with your own voice only.

And, uh, Finzar, if you’re watching this… please don’t sue me, dude.

get the hell out of here!

seems like no ai can say it like the original Now if you don’t like your own voice, you can choose one of the voices from the huge voice library here.

Anyways, so, the most asked question in this channel, after “how you edit your videos”, is “which voice are you using in ElevenLabs”. Well,

the voice you’re hearing right now, isn’t from the voice library or a cloned voice… It’s a custom voice! Most tutorials don’t mention this, but you can actually create an entirely new unique voice in ElevenLabs. It’s usually as realistic as the ones from the voice library, while being not available to everyone with one click. So it’s a great choice for creating

a brand “that doesn’t use this voice”. No one else can use it for their channel. Uhm,

well, if they don’t clone your voice with the cloning feature, but you get my point, right?

In the ‘voices’ tab, click ‘create a voice’, ‘voice design’. When I created my voice a few years ago, this feature was just a single button that you can click, and hope for a good voice.

But now, you can describe the voice in your mind however you want. Like “a young man in his 20s, american accent, speaking with quirky, but charismatic style. speaks with a lot of emotions and variances in his speech, changes the tone and speed continuously throughout the speech”. I

don’t know, something like that. And let’s reduce the guidance a bit to see what it comes up with.

Damn, can I change my voice to that? No? Okay…

Now, if you found a voice that works for you, we can move on to… Generating the actual voiceover.

At this point, you have a nice voice, and you know what realistic voice sounds like. So you can just copy your script, paste it to Elevenlabs, and hit ‘generate’ wait, hold on for a second. Of course,

you can do it. You may even get a pretty good result from it as well. But if you do that, after listening to it for a few minutes, something will feel off. And the viewers gonna feel it as well. I don’t know what it is, but it’s there in all AI voices. It’s like

they follow a pattern that just… gives it away… And once you notice it, you can’t unnotice it.

So let me show you how I generate my voiceovers. It takes much more time, but it’s the only way I found to get rid of that uncanny feeling. And it’s probably the secret sauce that makes you wonder how this ginger dude sounds so realistic. Alright, enough yapping, let’s get started.

First off, put the script down slowly.

We have to choose a voice model first. V3 is their latest model. It can do crazy stuff like making the voice giggle, whisper, or be excited. It sounds great, right? But when I tried it with my own voice… “it sounded like this”. Who the hell is that?! That’s like… my evil twin or something.

I even tried it with other voices, and every time, it slightly changes the voice itself. Except for a few voices that are designed for this model. So you can give them a try if you

itself. Except for a few voices that are designed for this model. So you can give them a try if you want. It’s still in the alpha stage, so hopefully it will be more stable in the future versions.

want. It’s still in the alpha stage, so hopefully it will be more stable in the future versions.

But it’s not really usable right now. So I use the V2 model instead for all my videos.

Unlike V3, this model has a few more sliders. Increase the speed, reduce the stability, and… turn of that annoying music! I’m trying to focus here, dude. Thank you.

Drag the similarity to something like 70%, and increase the style exaggeration.

Now select just a few sentences from your script, and paste it to Elevenlabs. What

did we needed for a realistic voice? Changing tone and speed, pauses, emphasis, and writing like a human. I assume you already nailed the last part. So let’s focus on the other three.

Using 3 dots makes the voice sound disappointed, or confused. To make it more excited, use exclamation marks. It always works. If it doesn’t, use more exclamations!

We’ll play with the pauses in the editing, but you can try 3 dots or commas to make them a bit more natural.

And for emphasizing important words, write them in all caps.

Another example, this is how it sounded like before making the changes.

“Say my name. You’re Isaac. You’re goddamn right” Pretty normal. And this is how it sounds like after making the changes.

Pretty normal. And this is how it sounds like after making the changes.

“Say my NAME! You’re ISAAC… You’re GODDAMN RIGHT!!”

Sounds much better, but if you’re not happy with how the result turned out, you can generate it again a few more times.

“you’re goddamn right. you’re goddamn right. you’re goddamn right” But why are we generating a few sentences at a time? Couldn’t we do the entire script at once?

Well, we could, but we shouldn’t. One reason is, if it generates a bad take, we can easily regenerate it without wasting too much credits. But the second reason? This is the secret sauce.

If you realized, each time you hit generate, “the tone changes slightly. the tone changes slightly. the tone changes slightly” even though we input the same text. So, we’re

slightly. the tone changes slightly” even though we input the same text. So, we’re

taking advantage of that. For each generation, you get 2 extra regenerations for free. Use

those free generations, and download them as well. You’re gonna see why in the editing part.

With every batch we generate, we get a slightly different tone and speed. That’s just like how normal humans sound like. So when combined together in the edit, it makes the final result sound more natural and fixes that uncanny feeling we talked about earlier.

After knowing these, it just takes a few hundred more generations, and you get this. Yes,

it takes ages, but it’s like cooking. Small changes make a huge difference.

Now put them all in a single folder, select them all, and rename. That’s much more organized.

Now we can move on to the next step.

--- I’m gonna show this in Premiere Pro, but you can do it in any editing software that has a timeline.

Import the voiceovers you generated to Premiere Pro, and start choosing the best takes. Don’t worry about timing or anything else at this point. Just

best takes. Don’t worry about timing or anything else at this point. Just

line the voiceover based on your script. If you downloaded a few versions of the same sentence, listen to all, and choose the one that fits the context better.

But the real magic, happens when you chop some parts from each generation, and combine them together. For this sentence, I generated 3 versions. Let’s listen to each one.

But the real magic, happens when you chop some parts from each generation, and combine them together.

But the real magic, happens when you chop some parts from each generation, and combine them together.

But the real magic, happens when you chop some parts from each generation, and combine them together.

Now they sound good on their own, but to make it more human-like, I took the best sections from each version, and got the final result.

But the real magic, happens when you chop some parts from each generation, and combine them together.

Once you got a single long straight line, start cutting the pauses. Use the second line to start the next sentence right after the last one ends. But don’t forget to add pauses and intentional spaces for the important parts of your video.

Now, even though Elevenlabs had a speed slider in its editor, you may realize just now that your voiceover is too slow, or too fast. Don’t worry, it happens… I mean I’m not the only one, right? Premiere Pro allows you to change the speed, but it turns you into a chipmunk if you

right? Premiere Pro allows you to change the speed, but it turns you into a chipmunk if you do that. Oh, there’s an option to ‘keep the pitch’. How’s that sound? Oh great,

do that. Oh, there’s an option to ‘keep the pitch’. How’s that sound? Oh great,

it distorts the voice now. Is it really that hard to make a speed changer, Adobe?

Fine, I’ll do that myself… Great, the voiceover is almost ready. But we can make it sound more professional. Here’s something

new I learned for this video. Take the parametric equalizer effect, add a high-pass filter, increase the h value, find echo-y frequencies, drag them down, take this slider, find a point where… Or, just drag and drop the preset I made! I also added a few fun voice effects and other useful stuff as well for you to try out. So feel free to download the link in the description.

Well, I think we’re done here. But wait a minute… What does youtube say? Does it allow using AI voices? Even more important, does it allow monetizing videos with AI voices?

Yes.

What, you expected a long list of explanations? Haha. I mean as long as you’re making an original content, not funny video compilations, yeah, you should be fine.

But should you be fine? I mean… should we all just start using AI voices now? What about, I don’t know… authenticity?

Wow… that’s… better than I could ever do… See? Told you it would work. Sounds great, man!

Yeah, but… it’s just… not me. What’s wrong, man? You’re becoming

a youtuber. Wasn’t that what you wanted? I don’t know, it feels like cheating. It

feels more like I’m becoming a fraud, dude. A fraud? You wanna know what’s fraud?

That… Silence… Having something worth saying, and keeping it locked up because you’re too scared of how it sounds. THAT’S fraud. Who says it should sound like you? As long as it’s your own thoughts, who cares how it sounds, huh? Come on, we don’t have time for this.

He was right. I was so obsessed with how I sound, I was forgetting the most important part. Actually saying something valuable… Yeah I mean I could end this video here, and call it a day, but that’s not all. I kept the best one for the last… I get it now. It’s a great solution if you don’t have a mic,

or if english isn’t your first language. I mean at least that’s progress, but you’re thinking too small, dude. You’re still thinking about replacement. Don’t you

see? Everything you learned, this isn’t just about finding a voice. There’s a bigger picture.

Man, talk in english, I don’t understand.

Okay, name a big youtuber. I don’t know, does MrBeast count?

Too cliche, but okay. When did his videos start to get viral?

I mean, he had over 10M subs in 2018. Yeah, but I mean, super viral. Look here. 2023.

Yeah that’s quite a jump. Now, when did he start dubbing his videos?

2023. Exactly! Ruhi Cenet.

Mark Rober. Joe Hattab. Anyone who started dubbing his videos to other languages, suddenly doubled, tripled their views and subscribers. Does it really make that much difference?

Think about it. There are 2.5 billion active youtube users in the world. Around half a billion of them speaks english. That’s less than 20%. If we just add up the next 5 most known languages to this, it makes up almost ha- Half of the entire youtube!

Do you see the potential? Sheet!

Well, that was the twist. AI voiceover isn’t just for narrating your scripts for you. It’s real

superpower is dubbing your videos. You can dub your videos, and get a massive audience. To try

it myself, I added just 2 languages to my last video, and the results were pretty clear… Oh, it’s actually blurry, heheh. Don’t worry, I’ll show it after showing how to dub your videos.

As an example, let’s take my last video. I already dubbed it into Spanish and Hindi, so let’s do Portuguese now.

Before we touch any AI, we need to separate the ingredients. After finishing your video, go to your editing timeline. Mute everything except the voiceover and export as an MP3. Next, do the opposite. Mute the voiceover, and export an MP3 with only the music and sound effects. You’ll see why in a second.

Now go to ElevenLabs and click the ‘Dubbing’ tab. Create a new dub. Select your original language, the target language, and upload that voiceover we just exported. Hit generate, and go make yourself a coffee. It takes a few minutes.

Once it’s done, download the file and drag it back into Premiere Pro. Here’s what you need to build here. The top, is your original native voice. We’re gonna use this as a guide for timing. Middle, is the new AI Portuguese voice. And

the bottom line, is for music and sound effects we exported earlier.

Line them up so the timing matches the visuals. Once it sounds good, mute the original English track, slap my preset on the Portuguese track to make it sound crispy, and export just the audio.

Now the only thing left, is to add it to your video. In YouTube Studio, click ‘languages’, and choose the video you just dubbed. Make sure youtube didn’t automatically dubbed to that language with their google translate voice. If so, just delete it from the menu here. Then click here, select the dubbed language, and upload your audio file.

Also, translate your video title and description with ChatGPT to reach the 99.1% purity.

And, you basically unlocked a new country! Keep repeating it for other languages as well, and you tripled your potential audience size.

Now as promised, here’s what it looked like in the analytics tab… So… yeah. It doesn’t look that impressive, does it? I mean, where’s the millions of views? Well,

So… yeah. It doesn’t look that impressive, does it? I mean, where’s the millions of views? Well,

hold on. If you look at the views by audio track instead… THIS, is how it looks like. Here’s Youtube’s auto-dubbed Hindi track. And THIS,

is where I added the Elevenlabs version. That’s what I’m talking about.

And this is the difference in spanish dub.

Wow… that’s… thousands of people… who watched my video! People who

would have seen none of it if I had listened to that voice in my head.

Turns out, Steven was right…