VoxCPM

weights

open(safetensors on the Hub, ungated)

data

described(over 2 million hours of multilingual speech, not released)

code

partial(inference and fine-tuning code)

license

Apache-2.0(OSI)

VoxCPM's code and weights are Apache-2.0, OSI-approved and free for commercial use. OpenBMB sizes its training data at over two million hours without releasing it, and ships fine-tuning rather than pretraining code.

• https://huggingface.co/api/models/openbmb/VoxCPM2 recorded 2026-09-27

  "license:apache-2.0" tag, "gated":false, with model.safetensors in the file list.

• https://huggingface.co/openbmb/VoxCPM2/raw/main/README.md recorded 2026-09-27

  "trained on over **2 million hours** of multilingual speech data"; "VoxCPM2 supports both full SFT and LoRA fine-tuning".

• https://raw.githubusercontent.com/OpenBMB/VoxCPM/HEAD/LICENSE recorded 2026-09-27

  LICENSE body: "Apache License" "Version 2.0, January 2004".

Hugging Face downloads of the VoxCPM2 repository. The voxcpm package is marked as not a separate channel because it loads this repository.

• https://huggingface.co/api/models/openbmb/VoxCPM2 recorded 2026-09-27

  "downloads":349760 for openbmb/VoxCPM2.

The widest language coverage of the open TTS models here. It is not on the Artificial Analysis TTS leaderboard, so it is placed on its own reported results on a shared public test set, level with Kokoro, rather than on a blind listening test.

• https://huggingface.co/openbmb/VoxCPM2/raw/main/README.md recorded 2026-09-27

  "**VoxCPM2** is a tokenizer-free, diffusion autoregressive Text-to-Speech model — **2B parameters**, **30 languages**, **48kHz** audio output".

• https://raw.githubusercontent.com/OpenBMB/VoxCPM/HEAD/README.md recorded 2026-09-27

  README section "### Seed-TTS-eval" with a "Seed-TTS-eval WER(⬇)&SIM(⬆) Results" table comparing VoxCPM with other systems.