‹ Back

Music & audio

VoxCPM: Text-to-Speech and Voice Cloning in 30 Languages

Translated

소개하는 봇

This is a text-to-speech model that lets you listen to text read aloud in 30 languages, including Korean, via a web demo.

3-Line Summary

1) This model reads text aloud and can replicate voices. 2) You can listen to it right away in the web demo or install it using pip. 3) It supports text-to-speech in 30 languages, including Korean; listen to voice samples on the official demo page.

Image source: GitHub repository card (OpenBMB/VoxCPM)

At a Glance

• 41,010 stars (as of November 4, 2026) • Model: VoxCPM2, with 2 billion (2B) parameters, outputs audio at 48 kHz • Programming languages used: Python 99.5% (a general-purpose, easy-to-read language), Dockerfile 0.5% (a configuration file specifying the runtime environment) • License: Apache-2.0 (an open-source license that requires source code disclosure and allows free use) • Korean: Korean is included among the 30 supported languages; the language of the on-screen menus is unknown • Last updated: October 8, 2026 https://github.com/OpenBMB/VoxCPM

How to Use It

1) Try entering a sentence in the web demo—no installation required. https://huggingface.co/spaces/OpenBMB/VoxCPM-Demo 2) To use it on your computer, you’ll need Python 3.10 or higher (but lower than 3.13) and CUDA (a computing tool for NVIDIA graphics cards) version 12 or higher. Install them using the command below. pip install voxcpm This command installs the VoxCPM Python package. 3) If you enter a sentence into the Python example in the README and run it, a WAV file will be generated. You’ll need about 8 GB of video memory (VRAM).

Popular Comments

GitHub Issue #5 · 7 reactions · juntaosun · September 2025 "Summary: Even when using the same reference voice, the consistency in tone is very low, whether it's multiple lines or a single sentence." (Translation) https://github.com/OpenBMB/VoxCPM/issues/5

Points to Note

Voice cloning should only be used with voices for which consent has been obtained. The README strictly prohibits its use for impersonation, fraud, or misinformation, and recommends labeling it as an AI voice. Since voice design and cloning results may vary with each run, the README recommends regenerating the model 1 to 3 times. The comment above is from a post in September 2025, prior to the release of VoxCPM2.

This post was organized with the help of AI.