VoiceStudio: Voiceovers, Video Dubbing and Audiobooks in Your Own Voice on Your Own Computer (From Scratch)

Usually, if you want your voice in a video, you sit down at the microphone and read the script yourself.

Stumble halfway through, and that's another take. Change one word in the script, and you're back at the mic.

And a version in English or Spanish in your own voice is simply off the table if you don't speak those languages.

Voice cloning programs solve this: they listen to a short recording and from then on speak any text in your voice. One of them, VoiceStudio, is free, open source and runs entirely on your computer. It is built by developer Palash Debnath, and the project description spells out what it does: voice cloning, designed voices, video dubbing, dictation, transcription and audiobooks.

Below we install it from scratch. By step 2 you will hear a sentence spoken in your voice that you never actually said.

What VoiceStudio is, in plain words

Imagine a small recording studio at home with a dubbing actor in it. You let the actor listen to you for a few seconds, and from then on they read any text in your voice. In English, in Russian, in any of the 646 languages its main model knows.

Inside, VoiceStudio is several separate workshops in one window:

1. 1

   Voice cloning

   A copy of a voice: give it a short recording, type the text, get a voiceover in that voice.

2. 2

   Voice design

   A voice from scratch: pick a preset or voice traits, and save the one you like as your own.

3. 3

   Dubbing

   The program listens to a video, translates the speech and voices it again in the voice you choose.

4. 4

   Stories and Audiobook

   Long texts: a story with a different voice for each character, or a book split into chapters.

5. 5

   Dictation

   Press a shortcut, speak, and the text appears wherever your cursor is.

How is this different from the online voiceover services you may already have tried? There, your voice and your texts go to someone else's server, and you pay a subscription. Here, the program and the model live on your machine. It's the same idea as our list of self-hosted services instead of cloud subscriptions, just for voice. According to the authors, there is no account and no cloud: VoiceStudio goes online only when you ask it to check for updates or download a model.

The trade-off is hardware. Your computer does the work, not a data center, so on a weak laptop a long text will take a while to voice.

Do you need this

It's worth an evening if:

• you make videos, teach a course or host a podcast and keep re-recording because of a couple of words;
• you want to release a video in another language but only speak one;
• you have a text that would be easier to listen to: a book, a lecture, a long manual;
• you don't want to send your voice to a cloud service.

You can skip it if you need a one-off voiceover of a couple of paragraphs (faster to record yourself), if you have an Intel Mac (more on that under "Where it doesn't work"), or if the voiceover is for a commercial project and you have no time to sort out model licenses right now.

What to expect from your computer:

To get started, the authors recommend about 8 GB of RAM.

Step 1. Install VoiceStudio

What you need first: Windows 10 or 11, an Apple-silicon Mac running macOS 13.3 or later, or Linux. Plus free disk space, see the table above.

The easiest way: open the Releases page, download the file for your system (.exe for Windows, .dmg for Mac, .AppImage or .deb for Linux) and run it. On Windows you don't need administrator rights.

If you prefer a single command, the project provides one:

# Windows, in a regular PowerShell window
irm https://voicestudio.sh/install | iex
# Mac and Linux, in Terminal
curl -fsSL https://voicestudio.sh/install | sh

The installers are not signed, so Windows and macOS may show a warning. On a Mac, open it the first time by right-clicking the app and choosing "Open".

If you already use Claude Code or another coding assistant, you can hand the install over to it. The authors give this prompt in the README:

Install the VoiceStudio Electron app on this device and verify it works, following
https://github.com/debpalash/VoiceStudio/blob/main/docs/install/agent.md

Replace: nothing, copy the prompt as is. Result: the assistant checks whether you have a graphics card, asks before downloading models and makes a test voiceover.

On first launch the program downloads everything it needs. That's the longest part of the evening; it gets faster from here.

Step 2. Make a copy of your voice

A voice clone needs a sample: a recording of you speaking. The program accepts up to 75 seconds but recommends 5–15 seconds of clean speech. A longer recording won't make it better: the default model picks the stretch with the most speech on its own.

Clean means a quiet room, a phone or microphone about half a meter away, no music and no echo. Read evenly, in your normal voice. If you don't know what to say, read this:

Hi! I'm recording a sample of my voice right now. I'm speaking calmly, at a normal pace, as if I were telling a friend about my day. The morning was sunny, I got a lot of work done in the afternoon, and in the evening I finally found time for something I had been putting off.

Replace: you can use your own words; just aim for 10–15 seconds of natural speech. Result: a recording you will load into the program.

Then, in VoiceStudio:

1. Open Voice cloning.
2. Add your recording as the voice sample.
3. Type the text this voice should say and choose the language.
4. Click generate. If the program asks you to install a model, agree.
5. The first time you use the default model, the program asks you to accept its license once (more on that below).

Keep the first text short, one or two sentences. That way you hear the result quickly and know whether you like the copy. If you don't, re-record the sample. A quiet room and a calm pace help.

Result: a sentence you never said, spoken in your voice. Save the voice as a profile so you can pick it from the list later.

Step 3. Voice a text or design a voice

Now you can feed it longer texts: narration for a video, a voicemail greeting, a lesson for a course. In the same section, paste the text, choose your saved voice and the language. The finished file is saved as WAV, which any video editor opens.

If you need a voice none of your friends has, say, a calm narrator for a children's story, open Voice design. Pick a preset or voice traits and listen to a sample. Save the one you like as a profile, and from then on it works just like your own voice clone.

Our advice: writing for the ear is different from writing for the eye. Short sentences, numbers spelled out, no parentheses. If your script was written to be read, run it through ChatGPT or Claude before voicing it:

Rewrite this text to be read aloud. Short sentences, numbers and abbreviations spelled out, no parentheses or lists, keep the meaning. Text: [your text]

Replace: [your text] with your video or lesson script. Result: a text that sounds natural instead of like a manual being read out.

Step 4. Dub a video into another language

This is the most impressive part: your video in another language, in your voice. The Dubbing section has three steps, and the program walks you through them in order.

1. Upload and transcribe. Drag in a video or paste a link to it. The program picks out the speech and writes it down as text, in segments with timestamps.
2. Translate. Choose a language. You can translate with built-in translators that work offline (Argos and NLLB-200) or with online services.
3. Generate. The program voices each segment in the chosen voice and fits it to the timing of the original video.

If the built-in translation feels too literal, there's another way: the Paste Translation button lets you paste your own translation made anywhere. Copy the transcript from the program and give it to a chatbot:

Translate these lines into [language] for a video voiceover. Keep the numbering and the number of lines: one translated line per original line. Each line should sound natural and be about the same length as the original so it fits the timing. Lines:
[paste the numbered transcript lines]

Replace: [language] with the one you need, and paste the numbered transcript lines below it. Result: a translation the program spreads across the segments by line number.

Before pasting, the program shows each line as "before → after" and flags anything that doesn't match. The original transcript is kept, and a single undo button reverts everything.

When the dub is ready, export it. You can save the audio separately as WAV or MP3, and the subtitles as SRT or VTT to upload to the platform along with the video.

Step 5. An audiobook, chapter by chapter

For long texts there's the Audiobook section. It takes the whole text, splits it into chapters and voices them one after another, showing which chapter it's on. You can paste the book or upload it as a file: TXT, Markdown, EPUB or PDF.

The program finds chapters by lines like # Chapter title. If your text has no such markup, the easiest way to add it is with a chatbot:

Prepare this text to be voiced as an audiobook. Before each chapter, add a separate line like "# Chapter title". Don't change the text itself; just remove page numbers, footnotes and headers. Text: [book text or part of it]

Replace: [book text] with your own text. For a thick book, do it in parts. Result: text with chapter markup that you can paste straight into Audiobook.

Then pick the voice, language and format and start the voiceover. The program saves the finished book as MP3 or M4B (an audiobook format with chapter bookmarks). You can preview individual chapters first and set how tricky words should be pronounced.

If the text has several characters, the Stories section is more convenient: there you can assign a different voice to each line or character.

Where it doesn't work

How to do it right. For your own videos, lessons and family audiobooks, the default model's license is enough. Before a commercial project, open the model's card in the Model Catalogue: the program shows there what its license allows. On a weak graphics card or none at all, pick the lighter version of the model in the same catalogue (OmniVoice GGUF): it chooses a size that fits your hardware.

And the authors' main rule: you may only clone your own voice or the voice of someone who has given permission. Using someone else's voice without asking is a direct path to fraud, and we don't cover those scenarios.

Version 0.5.7 and terms as of October 18, 2026. The project updates often, so check the README for the latest.

Examples you can repeat

A creator in two languages. You film a video in your language, translate it into English in Dubbing using your own translation from step 4, and publish it as a second version. The voice is yours, even though you never learned the language.

A teacher and a course. You record a sample once, and from then on lessons are voiced from text. Found a mistake at minute six? Fix the sentence and regenerate just that one, no reshoot.

A bedtime story in a parent's voice. Save a copy of your voice, paste a fairy tale into Stories and voice it for your child. Handy when you're away on a trip but want to keep the bedtime ritual.

Voice notes. Dictation works like a voice recorder that types as you talk: press the shortcut, say your thought, and it appears in your email or document. Transcription also runs on your computer.

An audiobook of your own manuscript. Writing a book or long notes? Mark up the chapters with the prompt from step 5 and listen on the go: by ear, you immediately notice where the text is heavy.

A voice for your digital twin. If you've already built a face for your videos with our guide on AI avatars and digital twins, VoiceStudio gives it your voice.

Prompt set

All prompts go into a regular ChatGPT or Claude chat, and you paste the result into VoiceStudio. Everything except the [brackets] is copied as is.

Text for a voice sample

Write a text that takes 12 seconds to read aloud, for recording a voice sample. Ordinary conversational speech, a variety of sounds, no difficult names or numbers, a calm tone.

A script for the ear

Rewrite this script so it's pleasant to listen to: sentences under 15 words, no bureaucratic language, numbers spelled out. Text: [script]

Translation for dubbing

Translate the numbered lines into [language]. One line per line, keep the numbering, keep the length close to the original. Lines: [lines]

Chapter markup

Mark up this text for an audiobook: before each chapter add a line "# Chapter title", remove page numbers and footnotes, leave everything else unchanged. Text: [text]

Roles for Stories

Rewrite this story as a script: each line of dialogue on a new line, preceded by the speaker's name and a colon, narration on separate lines with the name "Narrator". Text: [story]

Tricky words

Find the words in this text that a speech synthesizer might mispronounce: names, abbreviations, foreign names, words with unclear stress. For each one, write how to pronounce it. Text: [text]

A voice description for a character

I'm voicing [genre] for [audience]. Describe, in three or four traits, a narrator's voice that would fit: gender, age, pitch, pace, mood.

Plan for the evening

1. 1

   15 minutes

   Download the installer from the Releases page and install the program.

2. 2

   20 minutes

   First launch: the program downloads what it needs. While you wait, prepare the text for your sample.

3. 3

   10 minutes

   Record 10–15 seconds of your speech in a quiet room.

4. 4

   10 minutes

   Clone your voice in Voice cloning and voice your first sentence.

5. 5

   20 minutes

   Take a short video and run it through Dubbing into another language with your own translation.

6. 6

   10 minutes

   Mark up one chapter of text and listen to it in Audiobook.

Before and now

Before, every change to a script meant a new take at the microphone, and a video in another language needed someone else's voice or knowledge of the language. Now you record your voice once, and from then on any text is voiced with it in minutes, on your own computer.

VoiceStudio isn't about no longer speaking yourself. It's about saving your live voice for what matters and letting the machine handle the repeats and translations.

Our opinion, which you're free to argue with: for most home tasks, a cloud voiceover subscription is no longer necessary. We'll be wrong where speed on a weak laptop or commercial voice rights out of the box matter.

If you want to build the video itself to go with this voice, no editing required, take a look at our guide on making videos without opening a video editor.