The argument in one line.
VoiceBox is a free, open-source, local-only AI voice studio that clones any voice from a short audio sample and generates speech across several TTS engines, replacing paid cloud voice-cloning subscriptions with a tool that runs entirely on your own machine.
Read if. Skip if.
READ IF YOU ARE…
- You want narration, a podcast host voice, or character voices without paying for a cloud voice-cloning subscription like ElevenLabs.
- You're comfortable installing desktop software and want your voice data to stay on your own machine instead of a third-party server.
- You make video essays, audiobooks, game dialogue, or multi-voice podcast content and need more than one distinct AI voice in the same project.
SKIP IF…
- You need a hosted API or cloud service you can call from code — VoiceBox is a single-machine desktop app, not a service.
- You don't have a dedicated GPU and need studio-grade generation speed — CPU-only rendering is noticeably slower.
The full version, fast.
VoiceBox is a free, open-source, local-first AI voice studio: it clones a voice from a short recording and generates speech through seven different text-to-speech engines, all running on your own computer with no cloud uploads or subscription fees. The video installs the Windows/Mac/Linux desktop app, records a voice sample, transcribes it, and creates a reusable voice profile, then compares engines like Qwen3-TTS, Chatterbox, and Kokoro for quality, language coverage, and speed. It goes on to build a multi-voice conversation in the built-in Stories editor and layers on audio effects plus GPU acceleration. Clean input audio and picking the right engine per use case matter more than any single setting.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:27
01 · Intro
Cold open using a cloned AI voice sample, then the host previews the promise: clone any voice, generate speech in 20+ languages, and build multi-voice content, all free and local.
00:27 – 02:18
02 · What Is VoiceBox? (Features Overview)
Local-first architecture explained: voice cloning from a few seconds of audio, seven TTS engines, a multi-track Stories editor, system-wide dictation hotkey, and cross-platform support.
02:18 – 04:30
03 · Download & Install (Windows / Mac / Linux)
Downloading the installer from voicebox.sh, picking the right build per OS/chip, and clicking past Windows Defender / macOS Gatekeeper warnings that are normal for new open-source apps.
04:30 – 07:30
04 · Creating Your First Voice Profile
Recording a clean voice sample, transcribing it with a built-in Whisper model, and saving it as a reusable named voice profile.
07:30 – 12:46
05 · Generating AI Speech (Engine Breakdown)
Typing text for the cloned voice to say and comparing output across engines: Qwen3-TTS for multilingual quality, Chatterbox for language coverage, Chatterbox Turbo for speed and emotion tags, Kokoro for instant low-power generation.
12:46 – 14:36
06 · Building Multi-Voice Stories
Using the Stories editor's drag-and-drop timeline to alternate between two cloned voices and export a single stitched, timed conversation file.
14:36 – 17:13
07 · Audio Effects & Pro Tips
Applying built-in effect presets (Robotic, Radio, Echo Chamber, Deep Voice), enabling GPU acceleration for faster generation, and five recording tips for cleaner clones.
17:13 – 18:25
08 · Final Thoughts & Where to Go Next
Recap of everything covered, pointer to the docs site and GitHub repo, and a close-out subscribe ask.
Lines worth screenshotting.
- VoiceBox clones a usable voice profile from as little as 20-30 seconds of clean audio, with no cloud account required.
- The app runs seven different TTS engines locally, including Qwen3-TTS, Chatterbox, Chatterbox Turbo, LuxTTS, and Kokoro, and you can switch engines per generation.
- Chatterbox Turbo supports inline emotion tags directly in the text you type, letting you direct tone without extra settings.
- Kokoro is small enough to generate speech instantly even on old hardware, at the cost of voice variety versus the larger engines.
- Background noise is described as the single biggest factor that degrades a cloned voice's quality, more than sample length or engine choice.
- A 20-30 second voice sample is the sweet spot the video recommends for producing a natural-sounding clone.
- The built-in Stories editor stitches multiple cloned voices into a timed, multi-speaker audio file, effectively a mini podcast production studio.
- GPU acceleration, CUDA on Windows/NVIDIA and Metal on Apple silicon, can speed up generation by 5x or more over CPU-only rendering.
- Each generation uses a different random seed, so regenerating the same line with the same voice produces a slightly different take every time.
- VoiceBox is MIT-licensed and built by an open-source community around developer Jamie Pine, with no paid tier anywhere in the workflow.
How to clone any voice for free without renting a service
AI VOICE CLONING
A single free desktop app can clone a voice from a short sample and generate multilingual speech across several AI engines, with clean input audio mattering more than any setting you'll tweak.
02What Is VoiceBox? (Features Overview)
- A genuinely local-first AI tool keeps your voice data on your machine, no third-party server ever touches the audio you record.
- One app running seven separate TTS engines means you can match the model to the job instead of settling for whatever a single cloud service ships.
- A built-in system-wide dictation hotkey turns an AI voice tool into a general typing-replacement utility, not just a novelty voice cloner.
03Download & Install (Windows / Mac / Linux)
- New open-source software will trigger Windows Defender / macOS Gatekeeper warnings by default, that's normal, not a sign the app is unsafe, especially when the source is public on GitHub.
- Picking the wrong build (Intel vs. Apple Silicon, for example) is the most common install snag, check your exact chip before downloading.
04Creating Your First Voice Profile
- A clean 20-30 second sample recorded in a quiet room outperforms a longer, noisier one, audio quality beats audio quantity for cloning.
- Automatic transcription of your sample is what actually trains the voice profile, not just the raw audio.
- A saved voice profile is reusable forever across every future project, so the setup cost is a one-time investment.
05Generating AI Speech (Engine Breakdown)
- Different TTS engines optimize for different things: Qwen3-TTS for multilingual quality and tone control, Chatterbox for the widest language coverage, Chatterbox Turbo for speed and emotion tags, Kokoro for near-instant generation on weak hardware.
- Every generation uses a different random seed, so regenerating the same line is a legitimate way to get a better take, not a bug.
- Comparing two engines side-by-side on the identical line is the fastest way to learn which one actually matches a real voice.
06Building Multi-Voice Stories
- A drag-and-drop multi-voice timeline turns single-line generations into a full scripted conversation, the difference between a novelty clip and usable dialogue content.
- Producing a multi-speaker audio scene that used to require studio time and multiple actors now takes minutes with two saved voice profiles.
07Audio Effects & Pro Tips
- Built-in effect presets let you reshape a cloned voice for genre-specific content without any external audio editor.
- GPU acceleration is the single biggest speed lever, up to 5x faster generation, and modern local AI tools increasingly auto-configure it for you.
- The recurring quality rules boil down to: quiet room, 20-30 second sample, natural pace, regenerate when it's off, and pick the engine for the job.
Terms worth knowing.
- Voice cloning
- Creating an AI model of a specific person's voice from a short audio sample so it can generate new speech in that voice.
- TTS engine
- Text-to-speech engine — the underlying AI model that converts written text into spoken audio in a given voice.
- Local-first
- Software designed to run and store all its data on the user's own computer, without sending anything to an external server by default.
- Qwen3-TTS
- A multilingual text-to-speech AI model known for high-quality output and the ability to follow instructions like tone or pace.
- Chatterbox / Chatterbox Turbo
- An open-source TTS engine covering 23 languages; the Turbo variant trades some quality for speed and supports inline emotion tags in the text.
- Kokoro
- A very small, fast open-source TTS model that ships with dozens of preset voices and can run on modest hardware.
- Seed (generation seed)
- A random value that introduces slight variation into each AI generation, so repeating the same request never produces an identical result.
Things they pointed at.
Lines you could clip.
00:20
“What if I told you that you could clone any voice, generate speech in over 20 languages, and build your own multi voice podcast or story, all for completely free and everything running directly on your own computer.”
16:13
“Background noise is the number one thing that tanks voice quality.”
18:17
“Everything we did today was completely free and ran entirely on your own machine. No data going to anyone else's servers.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphor
00:00Good morning, and thank you for taking a moment to listen to this recording. Today is a great opportunity to learn something new and make progress toward important goals.
00:10The weather outside is calm, and the quiet environment helps create a clear audio sample. What's good, everyone? Welcome back to the channel.
00:19So real talk. What if I told you that you could clone any voice, generate speech in over 20 languages, and build your own multi voice podcast or story, all for completely free and everything running directly on your own computer.
00:34No cloud uploads, no subscriptions, no data going anywhere you didn't tell it to go. That tool exists.
00:42It's called VoiceBox. It's fully open source. It's got over 30,000 stars on GitHub.
00:48And today, I'm going to walk you through everything from installing it to making your first production ready audio. Whether you're a content creator, a developer, a student, or just someone curious about AI voice tech, this tutorial is built for you.
01:03Let's get into it. Before we jump into the install, let me give you a quick breakdown of what VoiceBox actually is because this thing is more than just a voice cloner. VoiceBox is a local first AI voice studio.
01:17What does local first mean? It means everything runs on your machine. Your voice data never leaves your computer.
01:25No third party servers are touching your audio ever. Here's what it can do. It can clone a voice from just a few seconds of audio.
01:33You record yourself or upload a sample, and it creates a voice profile you can use forever. It can generate speech using seven different AI engines, including Quin three TTS, Chatterbox, Luxe TTS, and more.
01:48Each engine has different strengths, and you can switch between them per generation. It has a stories editor, a full multitrack timeline where you can set up conversations between different voices, kind of like a all inside the app.
02:04It supports dictation, meaning you can hold a hotkey anywhere on your machine, speak, and your transcript gets pasted directly into whatever app you're using.
02:13And it works on Windows, Mac, and Linux. No extra setup. No terminal commands.
02:19Just download and go. The project was built by Jamie Pine and the open source community, and it's completely free under the MIT license.
02:28Link to the GitHub repo and the official site are down in the description. Alright. Let's get this thing installed.
02:35Step one. Head to the website. Open up your browser and go to voicebox.sh.
02:41I'll also drop the link right in the description below so you can just click it. Once you're on the site, scroll down a little until you see the download section. You'll see options for different operating systems.
02:52Step two, pick your platform. If you're on a Windows machine, click on Windows 64 bit MSI.
02:59That's going to download an installer file, the kind that walks you through setup with just a few next, next, finish clicks. Super easy. If you're on a Mac with Apple silicon, that's any Mac with an m one, m two, m three, or m four chip.
03:15Click Apple Silicon ARM. If you're on an older Intel Mac, click the Intel option instead. Not sure which Mac you have?
03:24Click the Apple logo in the top left corner of your screen, then click about this Mac, and it'll tell you right there. For Linux users, VoiceBox doesn't have a prebuilt binary just yet, but the site has a full build from source guide at voicebox.sh/linux-install. That's in the description too.
03:44Step three, run the installer. Once your file finishes downloading, click on it to open it. On Windows, you might see a security prompt from Windows Defender saying it doesn't recognize the app.
03:54Don't panic. This is totally normal for new open source software. Click more info and then hit run anyway.
04:02The app is open source and safe. Anyone can read its code on GitHub. Walk through the installer, hit next, choose where you want to install it, and then hit install.
04:12When it's done, hit finish. On Mac, open the DMG file that downloaded, drag the voice box icon into your applications folder, then open it from there.
04:22If macOS warns you about an unverified developer, go to system settings, then privacy and security, scroll down, and click open anyway.
04:32Step four, launch VoiceBox. Once the install is done, open VoiceBox. The first time it loads, it'll set up a few things in the background.
04:41Give it a moment. And boom. You're in.
04:46Welcome to the VoiceBox interface. Take a second to look around. On the left side, you'll see the main navigation icons.
04:53In the center is where all the action happens. Let's start by creating our first voice. This is where the magic starts, so pay close attention.
05:01Step one, navigate to voices. On the left sidebar, look for the icon that looks like a microphone. That's the voices section.
05:08Click on it. You'll see a button in the top right corner of the screen that says new voice. Go ahead and click on that.
05:15Step two, choose your input method. If the app asks for the microphone access, just click allow. You can now see the create voice screen, and here is the dialogue with three options for getting your voice sample into the app.
05:27Option one is upload. If you already have an audio file, like an m p three or WAV, you can drag it in here.
05:35Maybe you have a recording of yourself, a voice over you recorded earlier, or even a clean audio clip of a voice you wanna work with. Option two is microphone. This lets you record directly inside the app right now.
05:47You get up to thirty seconds of recording time, and the app actually shows you a live waveform, so you know it's picking up your voice. Option three is system audio.
05:57This captures whatever audio is playing on your computer in real time. For this tutorial, let's go with the microphone option so we can do this together. Click on record from microphone.
06:08Step three, record your sample. Before you hit record, find a quiet spot.
06:12The cleaner your audio, the better your cloned voice will sound. No fans blowing. No background music.
06:18No one talking in the other room. When you're ready, click the record button. Speak naturally.
06:24Say your name. Describe what you're doing. Tell a short story.
06:27The goal is to get close to that thirty second mark if you can. Longer samples almost always produce better results. When you're done speaking, click stop.
06:36You'll see a preview of your recording. Step four, transcribe your audio. Right below your recording, there's a button that says transcribe.
06:45Click on that. The first time you do this, VoiceBox will download a whisper model to process your audio. This is a speech to text model that helps the app understand what you said.
06:54It'll only take a minute or two, and it only happens once. After transcription finishes, you'll see all your words printed out in a text box below. This text paired with your audio is what trains your voice profile.
07:06Step five, name your voice and create the profile. On the right side of the screen, you'll see a field to name your voice. Type in whatever you wanna call it, your name, a character name, whatever makes sense to you.
07:18In the personality section, you can describe who this voice is and how they talk. In the default engine, there are text to speech models such as Quen three TTS, Luxe TTS, and others.
07:31I'll just go with the no preference option as of now, then click create profile in the bottom right corner. Your voice profile is now saved. You can use it anytime across any project forever.
07:44Now let's make your voice say something it never actually said. This is the part that genuinely blows people's minds. Step one, go to the generate tab.
07:54Here you can see our cloned voice is available, and in the bottom, we can see a text box. This is where you type what you want the voice to say. Step two, select your voice.
08:04Here, just select the voice profile. So we only have one voice profile, and it's already selected. Step three, choose your TTS engine.
08:12Under this text box, you'll see an engine selector. This is where you pick which AI model generates the speech. Here's a quick breakdown so you can make the right choice.
08:22Quen three dash TTS is great for high quality multilingual speech. It even lets you give it instructions, like speak slowly or sound excited.
08:31I recommend starting with the 1,700,000,000 version for the best quality. Chatterbox is your go to if you need languages beyond English.
08:39It covers 23 languages. Chatterbox Turbo is super fast and supports emotion tags like and right inside your text, which is really fun.
08:49Kokoro is tiny, runs instantly even on old hardware, and comes with over 50 preset voices built in. I have tried a couple of TTS engines including QEN three TTS 1,700,000,000. For your first generation, you can start with QEN three or Chatterbox Turbo.
09:07Select that from the drop down. Step four, type your text and generate. In the big text box below, type out whatever you want your cloned voice to say.
09:17Keep it short for your first try, maybe a sentence or two. Here, you can also select the effects such as robotic, echo, or deep from the drop down. Let's start with no effects for now.
09:28When you're happy with your text, click the generate button. Depending on your computer's hardware, it may take a few seconds to a minute. The first time also downloads the model, so give it a moment.
09:39Once it's done, you'll see your generation appear below. Hit the play button and listen to that. That's your voice or any voice you cloned, saying words it never actually spoke.
09:50Pretty incredible. Right? So this output is from Qwen three.
09:54Hey. Did you hear about that new AI tool everyone's talking about? And this one is from Chatterbox Turbo.
10:00Yeah. It's called VoiceBox. I've been using it all week.
10:04I think the later one is way more similar to my original audio. Step five, clone any voice. Now let's get to the most exciting part, cloning a voice.
10:15Head over to the voices section on the left sidebar and click create voice. You'll see a few options. Go ahead and click upload audio, then choose an audio file from your computer.
10:27The cleaner and longer your sample, the better the result. No background music, no noise, just a clear voice recording. Once your file is uploaded, click the transcribe button.
10:38VoiceBox will process your audio and convert everything you said into text. You can also hit the play button to listen back to your original sample and confirm it uploaded correctly. Now over on the right side, give this voice a name.
10:52Then in the bottom right corner, click create profile. And just like that, VoiceBox has built a complete AI voice profile from your sample. You'll see it appear in your voices list instantly.
11:04Now here's where it gets wild. Select your new voice profile, then paste any text you want it to say into the input box below.
11:12Hit the generate speech button and give it a moment to process. Done. Hit play and listen to this.
11:18Good morning, and thank you for taking a moment to listen to this recording. Today is a great opportunity to learn something new and make progress toward important goals. The weather outside is calm, and the quiet environment helps create a clear audio sample.
11:32That is a fully AI generated voice
11:35cloned from a audio sample, generated in seconds, running entirely on this machine with zero cloud upload and subscription fee. And the process works exactly the same for any voice you have permission to clone.
11:48Your own, a character you created, or someone who has given you explicit consent to use their voice. Same steps every time.
11:57Incredible. Right? Let's keep going.
11:59Step six, export your audio. If you like what you hear, click the three dots next to your generation and hit export.
12:08You can save the file as a WAV or MP3 to your computer and use it however you want, in videos, presentations, podcasts, anything.
12:17If you wanna try again with a different energy or phrasing, just hit regenerate, and it'll create a fresh take. Each take uses a slightly different random seed, so no two generations are exactly the same.
12:30Okay. So single voice generation is cool, but the stories editor, that's where VoiceBox starts feeling like a full production studio.
12:38The stories editor lets you build multi voice conversations. Think podcast episodes, audiobook dialogues, character conversations for games, you name it. Step one, open the stories section.
12:51On the left sidebar, click the stories icon. It looks like a timeline or stacked layers. Step two, create a new story.
12:59At the top of the screen, click new story. Give it a name, something like my first dialogue, and hit create. Step three, add your first line.
13:08At the bottom of the editor, you'll see a text field for adding a new line. Type in the first thing you want a character to say. For example, hey.
13:16Did you hear about that new AI tool everyone's talking about? Below that text field, there's a drop down to pick which voice profile speaks this line. Select your first voice, then click generate.
13:27You'll see the line appear as a clip on the timeline below. Step four, add more lines with different voices. Now click the text field again and type the response.
13:37For example, yeah. It's called voice box. I've been using it all week.
13:41This time, select a different voice profile from the drop down. Maybe you have a second voice you created, or you can import a preset voice from Chatterbox. Set it and hit generate.
13:52Keep going. Add as many lines as you want alternating between voices. Each line becomes its own clip on the timeline.
14:00Step five, arrange and export. The timeline is drag and drop. You can reorder clips, adjust the timing between lines, and preview the whole conversation by hitting the play button at the bottom.
14:13When you're happy with it, hit the export button in the top right corner. VoiceBox stitches all your clips together into a single audio file, perfectly timed, clean, and ready to use. Honestly, producing a full dialogue like this used to take hours of studio time.
14:30Now you can do it in minutes for free on your laptop. Let's talk about a couple bonus features that'll level up your output. Audio effects.
14:39On the left sidebar, click into the effects section. Here you'll find a built in post processing chain with eight different audio effects you can apply to any voice. VoiceBox ships with four built in presets right out of the box.
14:53Robotic, adds that classic digital glitchy robot sound. What do you think about this new tool?
14:59Great for sci fi content. Radio, makes voices sound like they're coming through a walkie talkie or vintage broadcast.
15:06What do you think about this new tool? Echo chamber. Adds deep reverb, almost like you're speaking in a cave.
15:13What do you think about this new tool? Deep voice. Lowers the pitch and adds weight to any voice.
15:19What do you think about this new tool? You can also build your own custom effect chain. Drag in effects like reverb, delay, chorus, compressor, pitch shift, and more.
15:29Tweak the parameters, preview in real time, and save it as your own preset. You can even set a default effects chain per voice profile so your voices always sound exactly the way you want.
15:41GPU acceleration. If your generations are feeling a little slow, head into settings and look for the GPU option. If you have a dedicated graphics card, whether that's NVIDIA, AMD, or even Intel Arc, select it here.
15:56On Windows with an NVIDIA card, VoiceBox can automatically download and set up CUDA support from right inside the app. On Mac with an m series chip, it already uses metal acceleration by default. This can make your generations five times faster or more.
16:13Pro tips for best results. Alright. Before we wrap up, here are a few quick tips to get the cleanest output possible.
16:20Number one, record in a quiet room. Background noise is the number one thing that tanks voice quality. Even just closing a window or moving away from a fan can make a big difference.
16:32Number two, aim for twenty to thirty seconds in your voice sample. The more data the model has to work from, the more accurate and natural the clone will be.
16:41Number three, speak clearly and at a natural pace when recording your sample. Don't rush. Don't over enunciate.
16:48Just talk like you normally do. Number four. If a generation doesn't sound quite right, try regenerating it two or three times.
16:57Different takes can sound dramatically different. Number five, experiment with TTS engines.
17:03Quen three dash TTS might nail the tone for narration, but Chatterbox Turbo might sound more expressive for dialogue. Try a few and see what fits your content.
17:13And that's a wrap on your complete beginner's guide to VoiceBox. We covered downloading and installing the app on any platform. We created a custom voice profile from a live recording.
17:24We generated speech using the Quen three t t s engine. We built a multi voice dialogue in the stories editor, and we explored effects, GPU settings, and pro tips to get the best results.
17:37And the best part? Everything we did today was completely free and ran entirely on your own machine. No data going to anyone else's servers.
17:46If you want to explore even deeper, the full documentation is at docs.voicebox.sh, and the GitHub repository is linked in the description.
17:56The community over there is really active, and the developers are always dropping new features and TTS engine support. If this video helped you out, drop a like. It genuinely helps more people find this content.
18:09And if you want more free AI tool breakdowns like this, hit subscribe and tap the bell so you never miss a video. Got questions? Drop them in the comments.
18:19I read everything, and I'll answer what I can. I'm out. Peace, and I'll catch you in the next one.
The Hook
The bait, then the rug-pull.
The video opens with a cloned AI voice reading a flat, generic paragraph out loud, before the host's real voice ever appears, which turns out to be the entire pitch of the tutorial that follows: build a voice exactly like this one, for free, on your own machine.
Named ideas worth stealing.
13:39list
VoiceBox TTS Engine Guide
- Qwen3-TTS — high-quality multilingual speech, accepts tone instructions, recommended at the 1.7B parameter size
- Chatterbox — broadest language coverage at 23 languages
- Chatterbox Turbo — fastest option, supports inline emotion tags
- Kokoro — smallest and fastest, 50+ preset voices, runs on old hardware
The host's rule of thumb for picking a TTS engine based on what a given generation needs to optimize for.
Steal forany multi-language or multi-voice AI audio project deciding which local TTS model to default to
16:15list
5 Pro Tips for Cleaner Voice Clones
- Record in a quiet room — background noise is the number one quality killer
- Aim for 20-30 seconds of sample audio
- Speak clearly at a natural pace, don't over-enunciate
- Regenerate 2-3 times if a take sounds off
- Match the TTS engine to the content — narration vs. dialogue
The host's closing checklist for getting the best clone and generation quality.
Steal forany voice-cloning or AI dictation workflow
How they asked for the click.
VERBAL ASK
18:16subscribe
“If this video helped you out, drop a like. And if you want more free AI tool breakdowns like this, hit subscribe and tap the bell so you never miss a video.”
Soft, single end-of-video ask delivered after full value has already been given, no mid-roll interruption, no sponsor pitch.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
