#99th of 18 in Voice
voxcpm
2B-parameter text-to-speech model with voice design and voice cloning
Live demo ↗
(opens in a new tab)Documentation ↗
(opens in a new tab)Website ↗
(opens in a new tab)Repository on GitHub ↗
(opens in a new tab)
- Stars
- 39.3k
- License
- Apache-2.0
- Last commit
- Oct 2026
- Last release
- May 2026
- Language
- Python
6,057 on 25 Mar 202639,310 on 14 Oct 2026
From GitHub's daily star counts.
Checked on GitHub.
Overview
VoxCPM2 is a 2B-parameter, tokenizer-free TTS model built on a MiniCPM-4 backbone. It synthesizes 48kHz speech in 30 languages, creates voices from a text description, and clones voices from a short reference clip, optionally with a transcript for continuation cloning. It ships as a pip package with a Python API, CLI and web demo, and can be served through Nano-vLLM, vLLM-Omni or llama.cpp-omni.
Who it is for: Developers needing self-hosted multilingual TTS, voice design or voice cloning
Strengths
- Voice design from a text description, no reference audio needed
- 30 languages plus nine Chinese dialects, no language tag required
- Apache-2.0 weights and code, with SFT and LoRA fine-tuning support
- Serving options: Nano-vLLM, vLLM-Omni (OpenAI-style endpoint), GGUF on llama.cpp-omni
Weaknesses
- Standard install requires CUDA 12.0 or newer and PyTorch 2.5.0 or newer
- Python support limited to 3.10 through 3.12
- RTF about 0.3 on RTX 4090 with the default PyTorch path
- llama.cpp-omni path is slower: RTF about 1.76 on Apple M4 Pro
What it needs
- GPU required
- Docker + Compose
- Needs PyTorch >= 2.5.0, CUDA >= 12.0, Python 3.10-3.12
- Models: VoxCPM2, VoxCPM1.5, VoxCPM-0.5B, GGUF (llama.cpp-omni)
- port 8808
Also in Voice
See all 18| Rank | Project | Score |
|---|---|---|
| 1 | voicestudioDesktop app for voice cloning, dubbing and dictation with local speech models | 81 out of 100 |
| 2 | whisper.cppC/C++ port of OpenAI Whisper for local speech-to-text transcription | 74 out of 100 |
| 3 | VoiceboxLocal voice studio for cloning, TTS, dictation and agent speech | 69 out of 100 |
| #4 | Speech-to-SpeechModular voice-agent pipeline exposed through the OpenAI Realtime API | 64 out of 100 |
| #5 | Kokoro-FastAPIOpenAI-compatible text-to-speech server for the Kokoro-82M model | 64 out of 100 |