Local text-to-speech models: open TTS by licence and hardware
The three best-rated open voices on the Artificial Analysis Speech Arena (25 October 2026) come with three different licences: Fish Drama 3 (commercial use with conditions), Breeze TTS 2 (non-commercial) and Ming-3.0-tts 0.5B (commercial use allowed). For a good voice on a laptop CPU, Kokoro 82M is Apache-2.0; for cloning, Qwen3-TTS, VoxCPM2 and Chatterbox all allow commercial use.
Written by Jakub Rusinowski · Last updated
The bar colour is the licence on the weights: the highest-rated open voices are not the ones you can ship without conditions.
The models
The licence column is the licence on the weights, read from each model card — not the badge on the code repository. Memory is what the maker or a named source publishes, not a measurement and not a fit verdict.
17 of 17 models
| Model | Hardware tier | Licence (weights) | Clones a voice | CPU only | Languages | Runtimes | Published memory | Speech Arena Elo (25 October 2026) |
|---|---|---|---|---|---|---|---|---|
| Kokoro 82M hexgrad · 82M | No GPU 8 GB GPU / 16 GB Mac | C Apache-2.0 Commercial use allowed | No | Yes | 8 | kokoro-onnx Kokoro-FastAPI LocalAI | ~2 GB (published by Pinggy Blog) single source under 2 GB VRAM, or ~4 GB RAM CPU-only via ONNX | 1,066 |
| Piper Open Home Foundation (OHF-Voice) | No GPU | ? per voice (code GPL-3.0) Not stated — read the model card | No | Yes | — | piper CLI Home Assistant Wyoming LocalAI | Not published | — |
| Pocket TTS Kyutai · 100M | No GPU | C CC-BY-4.0 — Attribution required. Commercial use allowed | Yes | Yes | 6 | pocket-tts (pip/uvx) ONNX INT8 export browser (WASM) | Not published | — |
| Qwen3-TTS Alibaba Qwen · 0.6B / 1.7B | 8 GB GPU / 16 GB Mac | C Apache-2.0 Commercial use allowed | Yes | — | 10 | vLLM-Omni LocalAI Transformers | ~4.54 GB (published by Pinggy Blog) single source 1.7B download 4.54 GB; 0.6B 2.52 GB | — |
| Chatterbox Resemble AI · 0.5B (Turbo 350M) | 8 GB GPU / 16 GB Mac | C MIT Commercial use allowed | Yes | — | 23 | pip chatterbox-tts LocalAI | Not published | 1,025 |
| VoxCPM2 OpenBMB · 2B | 16–24 GB GPU / 32 GB+ Mac No GPU | C Apache-2.0 Commercial use allowed | Yes | Yes | 30 | llama.cpp-omni (GGUF on CPU/Metal/CUDA/Vulkan) vLLM-Omni PyTorch | ~8 GB (published by OpenBMB (model card)) ~8 GB VRAM (card). | — |
| Step Audio EditX StepFun · 3B | 16–24 GB GPU / 32 GB+ Mac | ? not stated (code Apache-2.0) Not stated — read the model card | Yes | No | 6 | vLLM | ~12 GB (published by StepFun (model card)) 12 GB GPU memory is "a critical value"; 16 GB is safer (card). | 1,100 |
| Voxtral TTS Mistral AI · ~4B | 16–24 GB GPU / 32 GB+ Mac | NC CC-BY-NC-4.0 Non-commercial only | Yes | No | 9 | vLLM-Omni | ~16 GB (published by Mistral AI (model card)) A single GPU with at least 16 GB (card: BF16 weights). | 1,091 |
| Fish S2 Pro Fish Audio · ~4.4B (HF lists 5B) | Server GPU (rent) 16–24 GB GPU / 32 GB+ Mac | NC Fish Audio Research License Non-commercial only | Yes | No | 80 | SGLang-Omni vLLM-Omni | Not published | 1,110 |
| Breeze TTS 2 BreezeBlue (RESONIA) · 3B | 16–24 GB GPU / 32 GB+ Mac | NC BreezeBlue Research and Non-Commercial License Non-commercial only | Yes | No | 2 | PyTorch CLI Docker | ~12 GB (published by BreezeBlue (RESONIA) (model card)) ~7.7 GiB eager; 12 GB GPU minimum, 24 GB for the --fast-all path (card). | 1,230 |
| Higgs Audio V3 Boson AI · 4B | Server GPU (rent) 16–24 GB GPU / 32 GB+ Mac | NC Boson Higgs TTS 3 Research and Non-Commercial License Non-commercial only | Yes | No | 100 | SGLang-Omni | Not published | 1,045 |
| Magpie 357M NVIDIA · 364M (per card) | 8 GB GPU / 16 GB Mac | C NVIDIA Open Model License Commercial use allowed | No | — | 12 | NeMo | Not published | 1,065 |
| OmniVoice k2-fsa · 0.6B | 8 GB GPU / 16 GB Mac | NC CC-BY-NC Non-commercial only | Yes | — | 600 | vLLM-Omni | Not published | — |
| Fun-CosyVoice3 0.5B FunAudioLLM · 0.5B | 8 GB GPU / 16 GB Mac | C Apache-2.0 Commercial use allowed | Yes | — | — | vLLM-Omni | Not published | — |
| Fish Drama 3 Fish Audio · 35B total, 3B active (MoE) | Server GPU (rent) | C* Fish Audio Community License — Commercial use only below an annual-revenue threshold; never as a hosted service. Commercial use with conditions | Yes | No | — | SGLang-Omni vLLM-Omni | Not published | 1,248 |
| Ming-3.0-tts 0.5B inclusionAI (Ant Group) · 0.5B | 8 GB GPU / 16 GB Mac | C MIT Commercial use allowed | Yes | — | 7 | Ming-3.0-tts repository (PyTorch) | Not published | 1,191 |
| Phonon 31M Gradium · 31M | No GPU 8 GB GPU / 16 GB Mac | C Apache-2.0 Commercial use allowed | — | Yes | 1 | Ollama 0.40.11+ (Apple Silicon) | Not published | — |
No model matches these filters.
How to run it
Phonon in Ollama (Apple Silicon)
ollama pull phonon
ollama run phonon "Hello from Ollama."Ollama 0.40.11 or later, Macs only at launch; served at /v1/audio/speech. Ollama blog
Qwen3-TTS through llama.cpp
llama-tts -hf ggml-org/Qwen3-TTS-12Hz-1.7B-Base-GGUF -p "Hello world" --output out.wavFrom the llama.cpp TTS README; pass --tts-speaker-file to clone a voice. llama-server also serves POST /tts. llama.cpp TTS README
Kokoro, Chatterbox or VoxCPM2 in Python
pip install -q kokoro>=0.9.2 soundfile
pip install chatterbox-tts
pip install voxcpmOne line per model, each from its own card. Kokoro card · Chatterbox card · VoxCPM2 card
Sources: model cards on Hugging Face, the Open ASR Leaderboard, the Artificial Analysis Speech Arena and the Open TTS Leaderboard. Checked 25 October 2026.
Frequently asked questions
What is the best open text-to-speech model?
On the Artificial Analysis Speech Arena (25 October 2026) the highest-rated open-weight model is Fish Drama 3, at 1,248 Elo. Its licence is commercial use with conditions, so check it before you ship. Arena Elo measures listener preference for each maker’s own voices.
Which open TTS models can I use commercially?
Kokoro 82M, Pocket TTS, Qwen3-TTS, Chatterbox, VoxCPM2, Magpie 357M, Fun-CosyVoice3 0.5B, Ming-3.0-tts 0.5B, Phonon 31M have weights that allow commercial use. Read the model card before shipping: several licences forbid impersonation.
Can I clone a voice locally?
Yes. Qwen3-TTS, VoxCPM2, Chatterbox and Pocket TTS clone from a few seconds of reference audio. Clone only voices you have permission to use; several model licences explicitly prohibit impersonation.
Do I need a GPU for text-to-speech?
No. Kokoro, Piper and Pocket TTS run on a laptop CPU, and Gradium’s Phonon runs in Ollama on Apple Silicon. A GPU helps for the larger cloning models.
Why is the code Apache-2.0 but the model non-commercial?
Code and weights are licensed separately. Breeze TTS 2, for example, publishes its code under Apache-2.0 and its weights under a research and non-commercial licence. The weights are what bind you, which is why this page reads the model card.