VoxCPM

Name: VoxCPM
Author: VoxCPM

Free plan APILinuxMacSelf-hostedWeb

What VoxCPM is

VoxCPM is a text-to-speech system whose VoxCPM2 model generates expressive speech across 30 languages and lists nine Chinese dialects. It can create a voice from a written natural-language description, without reference audio, and supports voice cloning. The model produces 48 kHz audio, accepts 16 kHz reference audio, and lists WAV as an export format. A streaming generation API is available; the repository reports a real-time factor as low as about 0.3 on an NVIDIA RTX 4090. Users can fine-tune the model fully or with LoRA, with adaptation possible from 5–10 minutes of audio. Documented deployment paths include Python and CLI use, a local web demo, production serving, and on-device inference. Serving options include an OpenAI-compatible speech endpoint through vLLM-Omni and Nano-vLLM for concurrent requests. It is free and open source, with code and weights under Apache-2.0 and commercial use permitted. The documented Python setup requires Python 3.10 or later but below 3.13, PyTorch 2.5.0 or later, and CUDA 12.0 or later. Voice-design and controllable-cloning results can vary, and unsupported languages may need testing or fine-tuning. The project prohibits impersonation, fraud, and disinformation, and recommends labeling generated content.

Who it suits

VoxCPM suits developers and teams seeking a free, self-hosted or API-based speech system with multilingual generation, voice design, and cloning. Users should account for its documented installation requirements and variable voice-design results.

What it does well

Supports 30 languages and nine listed Chinese dialects
Voice design works without reference audio
Full and LoRA fine-tuning are available
Apache-2.0 code and weights permit commercial use

What to know first

Python setup requires CUDA 12.0 or later
Voice-design and cloning results can vary
Unsupported languages may need testing or fine-tuning

Limits

The documented Python installation requires Python 3.10 or later but below 3.13, PyTorch 2.5.0 or later, and CUDA 12.0 or later. Voice Design and controllable cloning results can vary between runs, and unsupported languages may need testing or fine-tuning.

Plans

Plan	Billed and limits	Price
Free & Open Source	No paid tiers or usage limits stated	Free

As voxcpm.com lists them, read 8 Oct 2026. The maker's own page has the final word.

Getting started

Open the VoxCPM website.
Choose web access or a documented deployment route such as Python, CLI, or on-device inference.
For the documented Python installation, use Python 3.10 or later but below 3.13, PyTorch 2.5.0 or later, and CUDA 12.0 or later.
Use the local web demo, production serving, or documented API option for speech generation.
For an OpenAI-compatible speech endpoint, use the vLLM-Omni serving option.

Verdict

VoxCPM combines multilingual speech generation with voice design, cloning, and multiple deployment routes. Its setup requirements and variable results are worth considering before adoption.

Questions

Is VoxCPM free?

Yes. The Free & Open Source plan is listed as free, with no paid tiers or usage limits stated.

Can I use VoxCPM commercially?

Yes. The project says its Apache-2.0 model weights and code are free for commercial use.

What platforms does it support?

Listed platforms are API, Linux, macOS, self-hosted, and web. The README also documents Apple Silicon MPS and C++ inference on CPU, Metal, CUDA, or Vulkan.

Does VoxCPM provide an API?

Yes. It has a streaming generation API. Its vLLM-Omni serving option exposes an OpenAI-compatible `/v1/audio/speech` endpoint.

What languages and audio formats does it support?

VoxCPM2 supports 30 languages and lists nine Chinese dialects. WAV is a listed export format; output is 48 kHz and reference audio is accepted at 16 kHz.

Who makes VoxCPM?

VoxCPM is made by VoxCPM, headquartered in Beijing, China.

Facts from VoxCPM's own pages

API access: Yes · voxcpm.com, 20 Sept 2026
API integration: Its vLLM-Omni serving option exposes an OpenAI-compatible /v1/audio/speech endpoint. · github.com, 8 Oct 2026
Audio output: The model outputs 48kHz audio and accepts 16kHz reference audio. · github.com, 8 Oct 2026
Commercial use: The project states its code and weights use the Apache-2.0 license and are free for commercial use. · github.com, 8 Oct 2026
Deployment: The repository documents Python and CLI use, a local web demo, production serving, and on-device inference. · github.com, 8 Oct 2026
Export formats: WAV · voxcpm.com, 20 Sept 2026
Fine-tuning: It supports full and LoRA fine-tuning, with adaptation possible from 5–10 minutes of audio. · github.com, 8 Oct 2026
Headquarters: Beijing, China · voxcpm.com, 28 Sept 2026
Integrations: Documented serving options include vLLM-Omni with an OpenAI-compatible speech endpoint and Nano-vLLM for concurrent requests. · github.com, 8 Oct 2026
Languages: VoxCPM2 supports 30 languages and lists nine Chinese dialects. · github.com, 8 Oct 2026
License and cost: The model weights and code use the Apache-2.0 license, and the repository says they are free for commercial use. · github.com, 8 Oct 2026
Limitations: Voice Design and controllable cloning results can vary between runs, and the project says unsupported languages may require testing or fine-tuning. · github.com, 8 Oct 2026
Maker: The site identifies Modelbest Technology Co., Ltd. and offers custom solutions through [email protected]. · voxcpm.com, 8 Oct 2026
Platforms: web,macos,linux,api,self_hosted · voxcpm.com, 20 Sept 2026
Platform support: The README documents Apple Silicon MPS support and C++ inference on CPU, Metal, CUDA, or Vulkan. · github.com, 8 Oct 2026
Product: VoxCPM2 is a tokenizer-free text-to-speech system that generates expressive multilingual speech. · github.com, 8 Oct 2026
Requirements: The documented Python installation requires Python 3.10 or later but below 3.13, PyTorch 2.5.0 or later, and CUDA 12.0 or later. · github.com, 8 Oct 2026
Responsible use: The project forbids using VoxCPM for impersonation, fraud, or disinformation and recommends labeling AI-generated content. · github.com, 8 Oct 2026
Streaming: It provides a streaming generation API, and the repository reports real-time factor as low as about 0.3 on an NVIDIA RTX 4090. · github.com, 8 Oct 2026
Support: The project links to Feishu, Discord, and its documentation wiki for community discussion and support. · github.com, 8 Oct 2026
Voice cloning: It supports controllable voice cloning from a short reference clip and continuation cloning with reference audio and its transcript. · github.com, 8 Oct 2026
Voice design: It can create a voice from a natural-language description without reference audio. · github.com, 8 Oct 2026