In this tutorial video, Helena Liu demonstrates how to obtain and run Microsoft’s open‑source VibeVoice text‑to‑speech model locally, bypassing the official version that was pulled from the internet due to deep‑fake concerns. She explains the history of VibeVoice—its initial release, rapid removal, and the community‑preserved GitHub copy—then walks through cloning the repository using Claude Code, installing either the 1.5 B or 7 B model, and launching a local server that supports up to four speakers, multilingual output, and custom voice cloning. Helena shows live audio samples, including multi‑speaker podcasts and English‑Chinese mixed text, and details the process of creating a personal voice clone from a 30‑second WAV sample. She compares the quality to paid services like ElevenLabs and NotebookLM, argues the model is comparable or superior, and highlights numerous use cases such as audiobooks, customer‑facing audio, and blog‑to‑audio conversion. The video concludes with a promotion of a free AI‑agent course and a business‑focused AI consulting call.
Chapters
Forget ElevenLabs. Microsoft just released a free voice cloning tool so good that they actually had to take it off of the market 3 weeks after it was released. Okay, here's the tea, guys. So over a year ago, Microsoft developed VibeVoice, but they were too afraid to release it because the voices that it was able to generate were too realistic and they were afraid it would flood the market with deepfakes. Now, last year they finally released the project and they open sourced the entire thing, but within weeks there were way too much concerns about deepfake, so Microsoft pulled the entire project off of the internet. But here's the thing though, a lot of community members made copies of the entire GitHub repository before Microsoft took it offline. And in this video, I'm going to show you how to copy that entire repository so that you can clone your voice and create AI audio without having to pay any subscription fees anymore because you can run everything locally on your computer. In addition to that, the Vibe Voice model can also allow you to have multiple speakers so you can create an entire AI podcast without having to go on to NotebookLM anymore, and you can customize all of the voices. All right, let's dive into it. So let's first take a look at the GitHub repository here. It was voted number 1, and when we scroll down here, we can actually see some of the voice samples. Let me just play a 10-second clip here for you guys. Hello everyone and welcome to the Vibe Voice Podcast. I'm your host. Now don't worry, listening to them— Okay, I really don't think that is any different from paid subscriptions like ElevenLabs. It is just as good, if not better. Now let's get this particular repository installed. Now it's important to note that you don't want to install the official Microsoft version because they have actually stripped out all of the text-to-speech functionality. So now you can only use it to transcribe audio. You can't use it to generate audio anymore. So I actually found a community member who saved the latest version of this project before Microsoft took away all of the text-to-speech functionalities. So what we're going to do is go on to Claude Code, and you want to use the desktop version. And once I'm in Claude Code, I'm going to start a new chat. And here is the line of command I'm going to put in. Basically, I'm going to call git to clone this particular project and install it onto my computer. And I will be right back after this repository is installed. Okay, so after this project is installed, the next thing I want to do is download the Vibe voice model onto my local machine. Now know that there are 2 models that you can install. I installed the smaller version, which is 1.5B. There is also a bigger model, but it will take up more space on your computer. And that is a 7B. So if you want the bigger model, change the line of command that I have written to 7B instead of 1.5B. I will include all the copy and paste codes that you need to put into your Claude code in the descriptions below. Okay, so I waited about 30 minutes or so, and then I got the VibeVoice model downloaded onto my local machine. And now I am just asking Claude to spin up a server for me to test it out. Okay, I just opened up my Chrome and I put in the server that Claude spun up here for me. And you can see here we can choose the number of speakers, we can get it up to 4, and we can choose the voice that we want for each of these speakers. There are about a dozen or so voices that are preloaded that comes with Vibe Voice. And over here there's a random example button. So if we just want to test out the AI audio, We can select this random example and then we can press Generate Podcast. And we usually wait a couple minutes for this is done, so I'll be right back. Okay, so our audio sample is finished. In this example, I'm using the Alice voice. So let's just listen to a couple of seconds and see what we think.
Hello and welcome.
Whoa, did you see that? Like there is the inclination and there is actually expression in the voices. So I'm pretty happy with it so far. Let's keep listening.
Today I'm going to tell you about VibeVoice, a groundbreaking text-to-speech model developed for generating natural, expressive speech. VibeVoice uses an—
and there's the pauses in between as well—
with a voice-native architecture combining autoregressive language modeling with diffusion-based audio generation. This allows it to produce remarkably natural-sounding speech with proper intonation and rhythm.
Pretty good, right? So next, what I want to do is try out their podcasting option where we have multiple speakers speaking at the same time, and we're also going to add in multiple languages into the text as well. All right. So VibeVoice can actually generate audio in multiple languages as well. This example has English and Chinese in it, and here is the audio output. Let's have a listen now. I'm just going to get it to a random place and play about 10, 15 seconds here for you.
If you want to say adding and subtracting three-digit numbers is a piece of cake for me, you can say 三位数的加减法对我来说小菜一碟。Right?
So you can see that this model can do other languages pretty well as well. Okay. So the use cases for this is just enormous. You can use it to have like a custom audio whenever a new customer buys. You can use this to translate all your blog posts into an audio.
Version.
You can use this for your audiobooks. I mean, the, the, like, the list goes on and on. Basically anything that you're currently using your ElevenLabs for, you can now replace it for free with something like Vibe Voice. Okay, I also want to just like give a shout out to Quinn, which is the Chinese model that's open source for voice generation as well. That one also works really well. Vibe Voice and Quinn are my 2 favorite open source AI audio models. Okay. And then here we can even get it to generate a whole podcast episode with multiple speakers. This is very similar to NotebookLM, except you can customize the voices. Like, you can clone your own voice into it, which I'll just show you in a second. But I'll just show you what the multi-speaker now sounds like here. Okay, so I'll play you, you know, a 10-15 second clip of these 2 speakers talking right here.
Okay.
All right.
Today we're getting into one of the biggest debates in all of sports. Who's the greatest basketball player of all time? I'm so excited to have Thomas here to talk about it with me.
Thanks so much for having me, Linda. You're absolutely right. This question always brings out some seriously strong feelings.
Okay, so let's get right into it.
All right, pretty good, right? Okay guys, now we're going to take Vibe Voice to the next level, and I'm going to show you how you can clone your own voice. So first, what you will need is at least a 30-second clip of yourself talking in a quiet room. How I did this was I just pulled out the Voice Memo app on my iPhone and I just started randomly talking for about 30 seconds or so. Then I sent that voice memo over to my computer. Now if you're using a MacBook, by default the voice memo will create a .m4v file, but in order for Vibe Voice to learn your voice, they need a .wav WAV file, but we can actually ask Claude Code to convert the file for us automatically, put it into the right folder on our desktop, and clone our voice. Now here is how I did that. I went into Claude and put in this prompt right here: convert my voice sample in the download folder and put it in the right Vibe, uh, voice folder and convert it to the WAV file, right? And then after it converted my voice, Then I told it to restart the server and add in the Helena voice as a choice option. So now I am back in the server, and now this time when I look on the dropdown, you can see there is the Helena voice, which is the custom voice that I have just cloned. And now let's just try this out. Let's get, um, the AI version of me to say something like, hi, my name is Helena. And this is my AI voice. Let's see if this sounds like me. Okay, let's generate the audio. This may take a minute or so, and I'll be right back once this is done. Okay, let's listen to the audio now. Hi, my name is Helena, and this is my AI voice. Let's see if this sounds like me.
Wow.
I would definitely say this is pretty damn close to the paid models at the moment. So this kind of blew my mind how much it did sound like me. I'm gonna try sending this to my mom and see if she can see the difference or not, because I think it's pretty realistic clone right here. Anyway, I hope you guys got a lot out of this video. If you learned something new, please like and subscribe. That will help out this channel so much. I also have a free course that will teach you how to build your own AI agents to take a lot of the repetitive work off of your plate so that you can generate more leads, get more sales, and increase your profitability. So make sure you check out that course below as well. And I will also include all of the instructions and prompts that I use to install VibeVoice in the descriptions as well so that you can just copy and paste everything and get this installed for yourself. Lastly, if you're a business owner and you're looking to implement AI across your entire business, build an AI operating system, book a free call with me, and I can't wait to see you very soon, guys. Bye for now.
