Hand this case to your agent and build your own version

A key section of this article is locked, and an agent opening the link cannot see it either. Unlock it first, then copy the full text for your agent in one click.

Run AI Voice Synthesis Free on Your Own Computer

An open-source case study: use VoxCPM2 to do voiceovers on your own computer. Type and get speech, design a voice with one sentence, or clone your own voice from one recording. 48kHz high audio quality, Apache 2.0 free for commercial use. Suited to short videos where you don't want to appear on camera, podcasts without equipment, and projects that need voiceover. Requires an NVIDIA GPU.

Machine translation. Read the original

The problem

Making short videos without wanting to appear on camera, making podcasts without equipment, and projects needing lots of voiceover: voice actors are expensive, and online tools require payment.

Who it's for

Short video creators who don't want to show their faces, podcast beginners, and projects needing voiceover and multilingual narration.

What you get

Generate 48kHz voiceovers for free on your own computer, design a voice with one sentence, or clone a voice from your own recording.

How the workflow runs

01 Copy the VoxCPM Skill from this site and give it to your Agent
02 The Agent checks your graphics card and sets up the environment
03 Type or describe a voice to generate speech
04 Clone with your own recording, listen to the result, then batch produce

Tools and accounts

VoxCPM2 (Apache-2.0)
Python + PyTorch
NVIDIA GPU (CUDA 12+)
Claude Code / Codex

Short videos where you don't want to appear on camera, podcasts without recording equipment, and projects that need voiceover can now be handled on your own computer with a free, open-source model: VoxCPM2.

What It Can Do

Type and get speech. Enter a passage of text and generate speech directly.

Design a voice with one sentence. No need to hire a voice actor or record a reference clip. Write a description like "a calm female voice in her thirties," and it generates a new voice matching that description.

Clone your voice from one recording. Give it a short recording of yourself, and it can learn your timbre, tone and phrasing, and even closely reproduce your speaking habits. For a closer match, add a verbatim transcript of that recording.

48kHz high audio quality. The output approaches studio quality, good enough for videos and podcasts.

Free for commercial use. The code and model are both Apache 2.0 licensed, so you can use and modify them for free, including in commercial projects.

It also supports 30 languages and multiple dialects, including Sichuanese, Cantonese and Northeastern Mandarin.

How to Get Started

The fastest way is to have your Agent set it up: open the VoxCPM Skill page on this site, click "Copy prompt," and paste it into Claude Code or Codex. It will check your computer, set up the environment, and then generate voices the way you want.

If you prefer to do it yourself, it runs in three lines:

pip install voxcpm soundfile

from voxcpm import VoxCPM
import soundfile as sf
model = VoxCPM.from_pretrained("openbmb/VoxCPM2", load_denoiser=False)
wav = model.generate(text="(沉稳的三十多岁女声)大家好，欢迎收听。")
sf.write("hello.wav", wav, model.tts_model.sample_rate)

Who It Is For

People making short videos who don't want to show their face or use their own voice.
People who want to start a podcast but have no recording equipment.
People whose projects need a lot of voiceover, narration, or multilingual versions.