VoxCPM

VoxCPM is an open-source text-to-speech model from OpenBMB; the README lists ModelBest and THUHCSI among the institutions behind it. The project calls it a tokenizer-free text-to-speech system. It reads text aloud, designs a new voice from a written description, or clones a voice from a reference clip. The repository holds several versions, including VoxCPM-0.5B, VoxCPM1.5 and VoxCPM2, and we list the whole series as one entry. The project says it supports 30 languages including Chinese and English, and you use it on your own machine through a Python API, the command line or a Gradio web interface. What sets it apart is that all three ways of making a voice sit in one open model you can run yourself.

Who it's for:Developers and creators who can set up Python and want to make voiceovers or design voices for their own work locally. Readers who don't write code won't be able to use it.

Before you start: You need Python 3.10 to 3.12 and PyTorch 2.5 or later, installed with pip install voxcpm; it runs on CPU, Apple Silicon or NVIDIA GPUs. The first run downloads the model files. The web interface listens on the whole network by default with no login, so we don't recommend the default setting; bind it to 127.0.0.1 instead. The project forbids using it for impersonation, fraud or disinformation and suggests labelling output as AI-generated, but it has no step to confirm that a cloned speaker agreed. We don't recommend cloning a real person's voice without their clear permission. The project says the code and models are licensed under Apache-2.0, which allows commercial use.

Based on the project's public documentation. Not tested by AI Inspiration World.