AI Voice Cloning
JarodMica/ai-voice-cloningReviewed by AI
Checked by RepoPicks Radar on
What is AI Voice Cloning?
AI Voice Cloning is a web page that runs on your own computer, built around the Tortoise TTS model, where you upload samples of a voice and have it read text you type. It can also train on your own recordings, transcribe them with Whisper or WhisperX, and pass the result through RVC to make it sound closer to the target voice. It needs an NVIDIA graphics card: on Windows you unpack a ready package and run start.bat, on Linux you build a Docker image. The author archived the project in 2025 and points to newer models instead.
AI Voice Cloning is free and open source (GPL-3.0) and can replace ElevenLabs and Respeecher.
How easy is it to install AI Voice Cloning?
D · terminal, several tools or a specific graphics card · Windows
- Windows D
- Linux D
Windows is the easier side and still a D: you need an NVIDIA card, then you unpack a 7-Zip package hosted on Hugging Face and run start.bat, which downloads most of the models on the first start, with more downloaded later during generation and training. Linux means Docker plus the NVIDIA Container Toolkit and building the image yourself with ./setup-docker.sh.
Coming soon: install in 1 click with the RepoPicks app.
AI Voice Cloning on GitHub
- Stars
- 774on Nov 10, 2026
- Downloads
- —all releases
- Growth
- —no gain in the last reading
Stars over time
Facts
- Runs on
- Windows, Linux
- You need to install
- Python, Docker, Graphics card, FFmpeg
- To build from source
- Python, Docker, Graphics card, FFmpeg
- Disk space
- 150 MB–450 MB · estimate
- Accounts or keys
- None
- Cost
- Free · GPL-3.0; everything runs on your own machine, so the cost is the hardware and the electricity.
- License
- GPL-3.0 (Copyleft)
- Can replace
- ElevenLabs, Respeecher
- Last change on GitHub
- Jun 9, 2025
AI Voice Cloning risk signals (9)
- Attentionstale releases
Last release on 2024-04-23 (929 days ago)
automatic check
- AttentionDependencies with known vulnerabilities
Pinned dependencies with known vulnerabilities (OSV): gradio: GHSA-26jh-r8g2-6fpr, GHSA-279j-x4gx-hfrh, GHSA-37qc-qgx6-9xjv; transformers: GHSA-282v-666c-3fvg, GHSA-29pf-2h5f-8g72, GHSA-37mw-44qp-f5jm; starlette: GHSA-2c2j-9gv5-cj73, GHSA-82w8-qh3p-5jfq, GHSA-86qp-5c8j-p5mr
automatic check
- Attentionclones the voice of real people
The whole purpose is copying a voice from samples, and the README's RVC section tells you to bring "a **trained** RVC .pth file that you get from RVC or online" and drop it in `models/rvc_models/`. Making a real person appear to say words they never said can break voice and likeness rights, and impersonation laws in several countries.
AI review
- Attentioninstalls code from many outside sources
Dockerfile pulls code and model packs straight from third-party addresses while installing: `curl -L -o /tmp/rvc.zip https://huggingface.co/Jmica/rvc/resolve/main/rvc_lightweight.zip`, `pip install git+https://github.com/liyaodev/fairseq`, `git+https://github.com/JarodMica/rvc-tts-pipeline.git@lightweight`, `git+https://github.com/m-bain/whisperx.git` and a clone of `pyfastmp3decoder`. None of these are pinned to a fixed version, so what you get depends on the day you install.
AI review
- Attentioncompiled binary committed to the repository
The top-level file list contains `deepspeed-0.8.3+6eca037c-cp39-cp39-win_amd64.whl`, a pre-compiled Windows Python package stored in the repository and installed by the setup scripts. Its contents cannot be read in the source.
AI review
- InfoRepository archived
The repository is archived (read-only, no longer maintained)
automatic check
- Infoready package hosted outside github
README Setup: "Head over to the releases tab and download the latest package on Hugging Face" — the Windows package is not a GitHub release asset, which is also why Radar's check found "no platform binaries" in releases v3.0, v2.0 and v1.0.
AI review
- Infoweb interface without a password
README, Building and Running in Docker: "Visit `http://localhost:7860` or remotely with `http://<ip>:7860`". The Gradio page has no login, so anyone who can reach that port can use the voices and files on the machine.
AI review
- Infoauthor points to newer projects
README: "I don't have any further plans to maintain this repo (Tortoise TTS) so I will be archiving it... There are better alternatives out there that work just as well or better than Tortoise: XTTS, GPT-SoVITS, F5TTS, StyleTTS2, Zonos, etc." The code is kept only "in a working state".
AI review
Evidence is shown in its original language.
What Radar could not check about AI Voice Cloning
- How much disk the models really take after the first start and after training, since they are downloaded as you go
- Whether the setup still completes today, with unpinned installs from GitHub and Hugging Face and two years without maintenance
- Whether an AMD card works through the setup-rocm or DirectML scripts, which the README does not cover
- How close the cloned voice gets, and how long generation and training take on a given graphics card
- Whether the known vulnerabilities in the pinned gradio, transformers and starlette versions can be reached in normal use