Repo of the Day
tarpetos/mimvox
Published: Nov 16, 2026
Open repository ↗Clone your own voice on your own computer: speak any text in 60+ languages and dub videos into another language in your voice. Fully local. Early version
Summary
Mimvox is a local desktop app for cloning a permitted voice, generating speech in roughly 60 languages, turning documents into audiobooks, and dubbing or re-speaking videos. It combines OmniVoice’s short-reference, zero-shot synthesis with optional RVC training for a closer timbre match, while keeping recordings, models, and generated speech on the user’s computer.
What it is useful for
Engineers can use Mimvox to produce multilingual voiceovers, prototype narration workflows, create chapter-based audiobooks from EPUB, DOCX, PDF, Markdown, subtitles, and other text formats, or replace a video’s dialogue while retaining its music and sound effects. A short recording can produce usable speech immediately; longer guided recordings can be trained to better match the speaker’s timbre.
The app is useful when audio must remain local. It has no account, cloud processing, or telemetry, and it works offline after the initial setup downloads the toolchains and models. A version check occurs at launch, but it can be disabled.
Several constraints are important. The project is explicitly an early hobby version with expected bugs, changing features, and possible setup failures. Intel Macs and ARM64 Windows and Linux systems are unsupported. CPU-only execution is possible but slow, while 8 GB or more of video memory is recommended. Setup requires roughly 34 GB of free disk, or 40 GB with the better Redub translation model. Voice cloning is limited to the user’s own voice or a voice whose owner has explicitly consented. The included model weights make the application personal and non-commercial only.
How engineers can use it
For normal use, download the installer for the supported platform from the latest release, launch Mimvox, select the required components, and run the in-app setup. Then either import an existing recording, complete a guided Quick, Standard, or Pro reading, or paste custom text for analysis. Press Speak for immediate zero-shot output or Train for a closer match, then export results as formats such as WAV, MP3, FLAC, or M4A.
For development on Mimvox itself, the README documents:
git clone https://github.com/tarpetos/mimvox.git
cd mimvox
./setup.sh
cd app && cargo run --release
Use setup.cmd instead of setup.sh on Windows. The script installs the required Rust and C/C++ toolchains, builds the Python environments, downloads models, and writes runtime.toml; it can be rerun and skips completed work. The repository also documents the app-to-engine JSON-lines protocol and data layout for engineers extending or integrating the system.