## ποΈ VoxCPM: Tokenizer-Free TTS for Context-Aware Speech Generation and True-to-Life Voice Cloning
[](https://github.com/OpenBMB/VoxCPM/) [](https://arxiv.org/abs/2509.24650) [](https://huggingface.co/openbmb/VoxCPM-0.5B) [](https://modelscope.cn/models/OpenBMB/VoxCPM-0.5B) [](https://huggingface.co/spaces/OpenBMB/VoxCPM-Demo) [](https://openbmb.github.io/VoxCPM-demopage)
<div align="center">
<img src="assets/voxcpm_logo.png" alt="VoxCPM Logo" width="40%">
</div>
<div align="center">
π Contact us on [WeChat](assets/wechat.png)
</div>
## News
* [2025.09.30] π₯ π₯ π₯ We Release VoxCPM [Technical Report](https://arxiv.org/abs/2509.24650)!
* [2025.09.16] π₯ π₯ π₯ We Open Source the VoxCPM-0.5B [weights](https://huggingface.co/openbmb/VoxCPM-0.5B)!
* [2025.09.16] π π π We Provide the [Gradio PlayGround](https://huggingface.co/spaces/OpenBMB/VoxCPM-Demo) for VoxCPM-0.5B, try it now!
## Overview
VoxCPM is a novel tokenizer-free Text-to-Speech (TTS) system that redefines realism in speech synthesis. By modeling speech in a continuous space, it overcomes the limitations of discrete tokenization and enables two flagship capabilities: context-aware speech generation and true-to-life zero-shot voice cloning.
Unlike mainstream approaches that convert speech to discrete tokens, VoxCPM uses an end-to-end diffusion autoregressive architecture that directly generates continuous speech representations from text. Built on [MiniCPM-4](https://huggingface.co/openbmb/MiniCPM4-0.5B) backbone, it achieves implicit semantic-acoustic decoupling through hierachical language modeling and FSQ constraints, greatly enhancing both expressiveness and generation stability.
<div align="center">
<img src="assets/voxcpm_model.png" alt="VoxCPM Model Architecture" width="90%">
</div>
### π Key Features
- **Context-Aware, Expressive Speech Generation** - VoxCPM comprehends text to infer and generate appropriate prosody, delivering speech with remarkable expressiveness and natural flow. It spontaneously adapts speaking style based on content, producing highly fitting vocal expression trained on a massive 1.8 million-hour bilingual corpus.
- **True-to-Life Voice Cloning** - With only a short reference audio clip, VoxCPM performs accurate zero-shot voice cloning, capturing not only the speakerβs timbre but also fine-grained characteristics such as accent, emotional tone, rhythm, and pacing to create a faithful and natural replica.
- **High-Efficiency Synthesis** - VoxCPM supports streaming synthesis with a Real-Time Factor (RTF) as low as 0.17 on a consumer-grade NVIDIA RTX 4090 GPU, making it possible for real-time applications.
## Quick Start
### π§ Install from PyPI
``` sh
pip install voxcpm
```
### 1. Model Download (Optional)
By default, when you first run the script, the model will be downloaded automatically, but you can also download the model in advance.
- Download VoxCPM-0.5B
```
from huggingface_hub import snapshot_download
snapshot_download("openbmb/VoxCPM-0.5B")
```
- Download ZipEnhancer and SenseVoice-Small. We use ZipEnhancer to enhance speech prompts and SenseVoice-Small for speech prompt ASR in the web demo.
```
from modelscope import snapshot_download
snapshot_download('iic/speech_zipenhancer_ans_multiloss_16k_base')
snapshot_download('iic/SenseVoiceSmall')
```
### 2. Basic Usage
```python
import soundfile as sf
import numpy as np
from voxcpm import VoxCPM
model = VoxCPM.from_pretrained("openbmb/VoxCPM-0.5B")
# Non-streaming
wav = model.generate(
text="VoxCPM is an innovative end-to-end TTS model from ModelBest, designed to generate highly expressive speech.",
prompt_wav_path=None, # optional: path to a prompt speech for voice cloning
prompt_text=None, # optional: reference text
cfg_value=2.0, # LM guidance on LocDiT, higher for better adherence to the prompt, but maybe worse
inference_timesteps=10, # LocDiT inference timesteps, higher for better result, lower for fast speed
normalize=True, # enable external TN tool
denoise=True, # enable external Denoise tool
retry_badcase=True, # enable retrying mode for some bad cases (unstoppable)
retry_badcase_max_times=3, # maximum retrying times
retry_badcase_ratio_threshold=6.0, # maximum length restriction for bad case detection (simple but effective), it could be adjusted for slow pace speech
)
sf.write("output.wav", wav, 16000)
print("saved: output.wav")
# Streaming
chunks = []
for chunk in model.generate_streaming(
text = "Streaming text to speech is easy with VoxCPM!",
# supports same args as above
):
chunks.append(chunk)
wav = np.concatenate(chunks)
sf.write("output_streaming.wav", wav, 16000)
print("saved: output_streaming.wav")
```
### 3. CLI Usage
After installation, the entry point is `voxcpm` (or use `python -m voxcpm.cli`).
```bash
# 1) Direct synthesis (single text)
voxcpm --text "VoxCPM is an innovative end-to-end TTS model from ModelBest, designed to generate highly expressive speech." --output out.wav
# 2) Voice cloning (reference audio + transcript)
voxcpm --text "VoxCPM is an innovative end-to-end TTS model from ModelBest, designed to generate highly expressive speech." \
--prompt-audio path/to/voice.wav \
--prompt-text "reference transcript" \
--output out.wav \
--denoise
# (Optinal) Voice cloning (reference audio + transcript file)
voxcpm --text "VoxCPM is an innovative end-to-end TTS model from ModelBest, designed to generate highly expressive speech." \
--prompt-audio path/to/voice.wav \
--prompt-file "/path/to/text-file" \
--output out.wav \
--denoise
# 3) Batch processing (one text per line)
voxcpm --input examples/input.txt --output-dir outs
# (optional) Batch + cloning
voxcpm --input examples/input.txt --output-dir outs \
--prompt-audio path/to/voice.wav \
--prompt-text "reference transcript" \
--denoise
# 4) Inference parameters (quality/speed)
voxcpm --text "..." --output out.wav \
--cfg-value 2.0 --inference-timesteps 10 --normalize
# 5) Model loading
# Prefer local path
voxcpm --text "..." --output out.wav --model-path /path/to/VoxCPM_model_dir
# Or from Hugging Face (auto download/cache)
voxcpm --text "..." --output out.wav \
--hf-model-id openbmb/VoxCPM-0.5B --cache-dir ~/.cache/huggingface --local-files-only
# 6) Denoiser control
voxcpm --text "..." --output out.wav \
--no-denoiser --zipenhancer-path iic/speech_zipenhancer_ans_multiloss_16k_base
# 7) Help
voxcpm --help
python -m voxcpm.cli --help
```
### 4. Start web demo
You can start the UI interface by running `python app.py`, which allows you to perform Voice Cloning and Voice Creation.
## π©βπ³ A Voice Chef's Guide
Welcome to the VoxCPM kitchen! Follow this recipe to cook up perfect generated speech. Letβs begin.
---
### π₯ Step 1: Prepare Your Base Ingredients (Content)
First, choose how youβd like to input your text:.
1. Regular Text (Classic Mode)
- β
Keep "Text Normalization" ON. Type naturally (e.g., "Hello, world! 123"). The system will automatically process numbers, abbreviations, and punctuation using WeTextProcessing library.
2. Phoneme Input (Native Mode)
- β Turn "Text Normalization" OFF. Enter phoneme text like {HH AH0 L OW1} (EN) or {ni3}{hao3} (ZH) for precise pronunciation control. In this mode, VoxCPM also supports native understanding of other complex non-normalized textβtry it out!
---
### π³ Step 2: Choose Your Flavor Profile (Voice Style)
This is the secret sauce that gives your audio its unique sound.
1. Cooking with a Prompt Speech (Following a Famous Recipe)
- A prompt speech provides the desired acoustic characteristics for VoxCPM. The speaker's timbre, speaking style, and even the background sounds and ambiance will be replicated.
- For a Clean, Studio-Quality Voice:
- β
Enable "Prompt Speech Enhancement". This acts like a noise filter, removing background hiss and rumble to give you a pure, clean voice clone.
2. Cooking au Naturel (Letting the Model Improvise)
- If no reference is provided, VoxCPM becomes a creative chef! It will infer a fitting speaking style based on the text itself, thanks to the text-smartness of its foundation model, MiniCPM-4.
- Pro Tip: Challenge VoxCPM with any textβpoetry, song lyrics, dramatic monologuesβit may deliver some interesting results!
---
### π§ Step 3: The Final Seasoning (Fine-Tuning Your Results)
You're ready to serve! But for master chefs who want to tweak the flavor, here are two key spices.
- CFG Value (How Closely to Follow the Recipe)
- Default: A great starting point.
- Voice sounds strained or weird? Lower this value. It tells the model to be more relaxed and improvisational, great for expressive prompts.
- Need maximum clarity and adherence to the text? Raise it slightly to keep the model on a tighter leash.
- Inference Timesteps (Simmering Time: Quality vs. Speed)
- Need a quick snack? Use a lower number. Perfect for fast drafts and experiments.
- Cooking a gourmet meal? Use a higher number. This lets the model "simmer" longer, refining the audio for superior detail and naturalness.
---
Happy creating! π Start with the default settings and tweak from there to suit your project. The kitchen is yours!
---
## π Community Projects
We're excited to see the VoxCPM community growing! Here are some amazing projects and features built by our community:
- **[ComfyUI-VoxCPM](https://github.com/wildminder/ComfyUI-VoxCPM)**
- **[ComfyUI-VoxCPMTTS](https://github.com/1038lab/ComfyUI-VoxCPMTTS)**
- **[WebUI-VoxCPM](https://github.com/rsxdalv/tts_webui_extension.vox_cpm)**
- **[PR: Streaming API Support (by AbrahamSanders)](https://github.com/OpenBMB/VoxCPM/pull/26)**
*Have you built something cool with VoxCPM? We'd love to feature it here! Please open an issue or pull request to add your project.*
## π Performance Highlights
VoxCPM achieves competitive results on public zero-shot TTS benchmarks:
### Seed-TTS-eval Benchmark
| Model | Parameters | Open-Source | test-EN | | test-ZH | | test-Hard | |
|------|------|------|:------------:|:--:|:------------:|:--:|:-------------:|:--:|
| | | | WER/%β¬ | SIM/%β¬| CER/%β¬| SIM/%β¬ | CER/%β¬ | SIM/%β¬ |
| MegaTTS3 | 0.5B | β | 2.79 | 77.1 | 1.52 | 79.0 | - | - |
| DiTAR | 0.6B | β | 1.69 | 73.5 | 1.02 | 75.3 | - | - |
| CosyVoice3 | 0.5B | β | 2.02 | 71.8 | 1.16 | 78.0 | 6.08 | 75.8 |
| CosyVoice3 | 1.5B | β | 2.22 | 72.0 | 1.12 | 78.1 | 5.83 | 75.8 |
| Seed-TTS | - | β | 2.25 | 76.2 | 1.12 | 79.6 | 7.59 | 77.6 |
| MiniMax-Speech | - | β | 1.65 | 69.2 | 0.83 | 78.3 | - | - |
| CosyVoice | 0.3B | β
| 4.29 | 60.9 | 3.63 | 72.3 | 11.75 | 70.9 |
| CosyVoice2 | 0.5B | β
| 3.09 | 65.9 | 1.38 | 75.7 | **6.83** | 72.4 |
| F5-TTS | 0.3B | β
| 2.00 | 67.0 | 1.53 | 76.0 | 8.67 | 71.3 |
| SparkTTS | 0.5B | β
| 3.14 | 57.3 | 1.54 | 66.0 | - | - |
| FireRedTTS | 0.5B | β
| 3.82 | 46.0 | 1.51 | 63.5 | 17.45 | 62.1 |
| FireRedTTS-2 | 1.5B | β
| 1.95 | 66.5 | 1.14 | 73.6 | - | - |
| Qwen2.5-Omni | 7B | β
| 2.72 | 63.2 | 1.70 | 75.2 | 7.97 | **74.7** |
| OpenAudio-s1-mini | 0.5B | β
| 1.94 | 55.0 | 1.18 | 68.5 | - | - |
| IndexTTS2 | 1.5B | β
| 2.23 | 70.6 | 1.03 | 76.5 | - | - |
| VibeVoice | 1.5B | β
| 3.04 | 68.9 | 1.16 | 74.4 | - | - |
| HiggsAudio-v2 | 3B | β
| 2.44 | 67.7 | 1.50 | 74.0 | - | - |
| **VoxCPM** | 0.5B | β
| **1.85** | **72.9** | **0.93** | **77.2** | 8.87 | 73.0 |
### CV3-eval Benchmark
| Model | zh | en | hard-zh | | | hard-en | | |
|-------|:--:|:--:|:-------:|:--:|:--:|:-------:|:--:|:--:|
| | CER/%⬠| WER/%⬠| CER/%⬠| SIM/%⬠| DNSMOS⬠| WER/%⬠| SIM/%⬠| DNSMOS⬠|
| F5-TTS | 5.47 | 8.90 | - | - | - | - | - | - |
| SparkTTS | 5.15 | 11.0 | - | - | - | - | - | - |
| GPT-SoVits | 7.34 | 12.5 | - | - | - | - | - | - |
| CosyVoice2 | 4.08 | 6.32 | 12.58 | 72.6 | 3.81 | 11.96 | 66.7 | 3.95 |
| OpenAudio-s1-mini | 4.00 | 5.54 | 18.1 | 58.2 | 3.77 | 12.4 | 55.7 | 3.89 |
| IndexTTS2 | 3.58 | 4.45 | 12.8 | 74.6 | 3.65 | - | - | - |
| HiggsAudio-v2 | 9.54 | 7.89 | 41.0 | 60.2 | 3.39 | 10.3 | 61.8 | 3.68 |
| CosyVoice3-0.5B | 3.89 | 5.24 | 14.15 | 78.6 | 3.75 | 9.04 | 75.9 | 3.92 |
| CosyVoice3-1.5B | 3.91 | 4.99 | 9.77 | 78.5 | 3.79 | 10.55 | 76.1 | 3.95 |
| **VoxCPM** | **3.40** | **4.04** | 12.9 | 66.1 | 3.59 | **7.89** | 64.3 | 3.74 |
## β οΈ Risks and limitations
- General Model Behavior: While VoxCPM has been trained on a large-scale dataset, it may still produce outputs that are unexpected, biased, or contain artifacts.
- Potential for Misuse of Voice Cloning: VoxCPM's powerful zero-shot voice cloning capability can generate highly realistic synthetic speech. This technology could be misused for creating convincing deepfakes for purposes of impersonation, fraud, or spreading disinformation. Users of this model must not use it to create content that infringes upon the rights of individuals. It is strictly forbidden to use VoxCPM for any illegal or unethical purposes. We strongly recommend that any publicly shared content generated with this model be clearly marked as AI-generated.
- Current Technical Limitations: Although generally stable, the model may occasionally exhibit instability, especially with very long or expressive inputs. Furthermore, the current version offers limited direct control over specific speech attributes like emotion or speaking style.
- Bilingual Model: VoxCPM is trained primarily on Chinese and English data. Performance on other languages is not guaranteed and may result in unpredictable or low-quality audio.
- This model is released for research and development purposes only. We do not recommend its use in production or commercial applications without rigorous testing and safety evaluations. Please use VoxCPM responsibly.
## πTO-DO List
Please stay tuned for updates!
- [x] Release the VoxCPM technical report.
- [ ] Support higher sampling rate (next version).
## π License
The VoxCPM model weights and code are open-sourced under the [Apache-2.0](LICENSE) license.
## π Acknowledgments
We extend our sincere gratitude to the following works and resources for their inspiration and contributions:
- [DiTAR](https://arxiv.org/abs/2502.03930) for the diffusion autoregressive backbone used in speech generation
- [MiniCPM-4](https://github.com/OpenBMB/MiniCPM) for serving as the language model foundation
- [CosyVoice](https://github.com/FunAudioLLM/CosyVoice) for the implementation of Flow Matching-based LocDiT
- [DAC](https://github.com/descriptinc/descript-audio-codec) for providing the Audio VAE backbone
## Institutions
This project is developed by the following institutions:
- <img src="assets/modelbest_logo.png" width="28px"> [ModelBest](https://modelbest.cn/)
- <img src="assets/thuhcsi_logo.png" width="28px"> [THUHCSI](https://github.com/thuhcsi)
## β Star History
[](https://star-history.com/#OpenBMB/VoxCPM&Date)
## π Citation
If you find our model helpful, please consider citing our projects π and staring us βοΈοΌ
```bib
@article{voxcpm2025,
title = {VoxCPM: Tokenizer-Free TTS for Context-Aware Speech Generation and True-to-Life Voice Cloning},
author = {Zhou, Yixuan and Zeng, Guoyang and Liu, Xin and Li, Xiang and Yu, Renjie and Wang, Ziyang and Ye, Runchuan and Sun, Weiyue and Gui, Jiancheng and Li, Kehan and Wu, Zhiyong and Liu, Zhiyuan},
journal = {arXiv preprint arXiv:2509.24650},
year = {2025},
}
```
Raw data
{
"_id": null,
"home_page": null,
"name": "voxcpm",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.10",
"maintainer_email": "OpenBMB <openbmb@gmail.com>",
"keywords": "voxcpm, text-to-speech, tts, speech-synthesis, voice-cloning, ai, deep-learning, pytorch",
"author": null,
"author_email": "OpenBMB <openbmb@gmail.com>",
"download_url": "https://files.pythonhosted.org/packages/0e/94/0fb57fb764d25a9dec0f5842b0799c19eaa88573780e4e84d94d59e3e031/voxcpm-1.0.5.tar.gz",
"platform": null,
"description": "## \ud83c\udf99\ufe0f VoxCPM: Tokenizer-Free TTS for Context-Aware Speech Generation and True-to-Life Voice Cloning\n\n\n[](https://github.com/OpenBMB/VoxCPM/) [](https://arxiv.org/abs/2509.24650) [](https://huggingface.co/openbmb/VoxCPM-0.5B) [](https://modelscope.cn/models/OpenBMB/VoxCPM-0.5B) [](https://huggingface.co/spaces/OpenBMB/VoxCPM-Demo) [](https://openbmb.github.io/VoxCPM-demopage)\n\n\n\n<div align=\"center\">\n <img src=\"assets/voxcpm_logo.png\" alt=\"VoxCPM Logo\" width=\"40%\">\n</div>\n\n<div align=\"center\">\n\n\ud83d\udc4b Contact us on [WeChat](assets/wechat.png)\n\n</div>\n\n## News \n* [2025.09.30] \ud83d\udd25 \ud83d\udd25 \ud83d\udd25 We Release VoxCPM [Technical Report](https://arxiv.org/abs/2509.24650)!\n* [2025.09.16] \ud83d\udd25 \ud83d\udd25 \ud83d\udd25 We Open Source the VoxCPM-0.5B [weights](https://huggingface.co/openbmb/VoxCPM-0.5B)!\n* [2025.09.16] \ud83c\udf89 \ud83c\udf89 \ud83c\udf89 We Provide the [Gradio PlayGround](https://huggingface.co/spaces/OpenBMB/VoxCPM-Demo) for VoxCPM-0.5B, try it now! \n\n## Overview\n\nVoxCPM is a novel tokenizer-free Text-to-Speech (TTS) system that redefines realism in speech synthesis. By modeling speech in a continuous space, it overcomes the limitations of discrete tokenization and enables two flagship capabilities: context-aware speech generation and true-to-life zero-shot voice cloning.\n\nUnlike mainstream approaches that convert speech to discrete tokens, VoxCPM uses an end-to-end diffusion autoregressive architecture that directly generates continuous speech representations from text. Built on [MiniCPM-4](https://huggingface.co/openbmb/MiniCPM4-0.5B) backbone, it achieves implicit semantic-acoustic decoupling through hierachical language modeling and FSQ constraints, greatly enhancing both expressiveness and generation stability.\n\n<div align=\"center\">\n <img src=\"assets/voxcpm_model.png\" alt=\"VoxCPM Model Architecture\" width=\"90%\">\n</div>\n\n\n### \ud83d\ude80 Key Features\n- **Context-Aware, Expressive Speech Generation** - VoxCPM comprehends text to infer and generate appropriate prosody, delivering speech with remarkable expressiveness and natural flow. It spontaneously adapts speaking style based on content, producing highly fitting vocal expression trained on a massive 1.8 million-hour bilingual corpus.\n- **True-to-Life Voice Cloning** - With only a short reference audio clip, VoxCPM performs accurate zero-shot voice cloning, capturing not only the speaker\u2019s timbre but also fine-grained characteristics such as accent, emotional tone, rhythm, and pacing to create a faithful and natural replica.\n- **High-Efficiency Synthesis** - VoxCPM supports streaming synthesis with a Real-Time Factor (RTF) as low as 0.17 on a consumer-grade NVIDIA RTX 4090 GPU, making it possible for real-time applications.\n\n\n\n\n\n## Quick Start\n\n### \ud83d\udd27 Install from PyPI\n``` sh\npip install voxcpm\n```\n### 1. Model Download (Optional)\nBy default, when you first run the script, the model will be downloaded automatically, but you can also download the model in advance.\n- Download VoxCPM-0.5B\n ```\n from huggingface_hub import snapshot_download\n snapshot_download(\"openbmb/VoxCPM-0.5B\")\n ```\n- Download ZipEnhancer and SenseVoice-Small. We use ZipEnhancer to enhance speech prompts and SenseVoice-Small for speech prompt ASR in the web demo.\n ```\n from modelscope import snapshot_download\n snapshot_download('iic/speech_zipenhancer_ans_multiloss_16k_base')\n snapshot_download('iic/SenseVoiceSmall')\n ```\n\n### 2. Basic Usage\n```python\nimport soundfile as sf\nimport numpy as np\nfrom voxcpm import VoxCPM\n\nmodel = VoxCPM.from_pretrained(\"openbmb/VoxCPM-0.5B\")\n\n# Non-streaming\nwav = model.generate(\n text=\"VoxCPM is an innovative end-to-end TTS model from ModelBest, designed to generate highly expressive speech.\",\n prompt_wav_path=None, # optional: path to a prompt speech for voice cloning\n prompt_text=None, # optional: reference text\n cfg_value=2.0, # LM guidance on LocDiT, higher for better adherence to the prompt, but maybe worse\n inference_timesteps=10, # LocDiT inference timesteps, higher for better result, lower for fast speed\n normalize=True, # enable external TN tool\n denoise=True, # enable external Denoise tool\n retry_badcase=True, # enable retrying mode for some bad cases (unstoppable)\n retry_badcase_max_times=3, # maximum retrying times\n retry_badcase_ratio_threshold=6.0, # maximum length restriction for bad case detection (simple but effective), it could be adjusted for slow pace speech\n)\n\nsf.write(\"output.wav\", wav, 16000)\nprint(\"saved: output.wav\")\n\n# Streaming\nchunks = []\nfor chunk in model.generate_streaming(\n text = \"Streaming text to speech is easy with VoxCPM!\",\n # supports same args as above\n):\n chunks.append(chunk)\nwav = np.concatenate(chunks)\n\nsf.write(\"output_streaming.wav\", wav, 16000)\nprint(\"saved: output_streaming.wav\")\n```\n\n### 3. CLI Usage\n\nAfter installation, the entry point is `voxcpm` (or use `python -m voxcpm.cli`).\n\n```bash\n# 1) Direct synthesis (single text)\nvoxcpm --text \"VoxCPM is an innovative end-to-end TTS model from ModelBest, designed to generate highly expressive speech.\" --output out.wav\n\n# 2) Voice cloning (reference audio + transcript)\nvoxcpm --text \"VoxCPM is an innovative end-to-end TTS model from ModelBest, designed to generate highly expressive speech.\" \\\n --prompt-audio path/to/voice.wav \\\n --prompt-text \"reference transcript\" \\\n --output out.wav \\\n --denoise\n\n# (Optinal) Voice cloning (reference audio + transcript file)\nvoxcpm --text \"VoxCPM is an innovative end-to-end TTS model from ModelBest, designed to generate highly expressive speech.\" \\\n --prompt-audio path/to/voice.wav \\\n --prompt-file \"/path/to/text-file\" \\\n --output out.wav \\\n --denoise\n\n# 3) Batch processing (one text per line)\nvoxcpm --input examples/input.txt --output-dir outs\n# (optional) Batch + cloning\nvoxcpm --input examples/input.txt --output-dir outs \\\n --prompt-audio path/to/voice.wav \\\n --prompt-text \"reference transcript\" \\\n --denoise\n\n# 4) Inference parameters (quality/speed)\nvoxcpm --text \"...\" --output out.wav \\\n --cfg-value 2.0 --inference-timesteps 10 --normalize\n\n# 5) Model loading\n# Prefer local path\nvoxcpm --text \"...\" --output out.wav --model-path /path/to/VoxCPM_model_dir\n# Or from Hugging Face (auto download/cache)\nvoxcpm --text \"...\" --output out.wav \\\n --hf-model-id openbmb/VoxCPM-0.5B --cache-dir ~/.cache/huggingface --local-files-only\n\n# 6) Denoiser control\nvoxcpm --text \"...\" --output out.wav \\\n --no-denoiser --zipenhancer-path iic/speech_zipenhancer_ans_multiloss_16k_base\n\n# 7) Help\nvoxcpm --help\npython -m voxcpm.cli --help\n```\n\n### 4. Start web demo\n\nYou can start the UI interface by running `python app.py`, which allows you to perform Voice Cloning and Voice Creation.\n\n\n\n## \ud83d\udc69\u200d\ud83c\udf73 A Voice Chef's Guide\nWelcome to the VoxCPM kitchen! Follow this recipe to cook up perfect generated speech. Let\u2019s begin.\n\n---\n### \ud83e\udd5a Step 1: Prepare Your Base Ingredients (Content)\n\nFirst, choose how you\u2019d like to input your text:.\n1. Regular Text (Classic Mode)\n- \u2705 Keep \"Text Normalization\" ON. Type naturally (e.g., \"Hello, world! 123\"). The system will automatically process numbers, abbreviations, and punctuation using WeTextProcessing library.\n2. Phoneme Input (Native Mode)\n- \u274c Turn \"Text Normalization\" OFF. Enter phoneme text like {HH AH0 L OW1} (EN) or {ni3}{hao3} (ZH) for precise pronunciation control. In this mode, VoxCPM also supports native understanding of other complex non-normalized text\u2014try it out!\n\n---\n### \ud83c\udf73 Step 2: Choose Your Flavor Profile (Voice Style) \n\nThis is the secret sauce that gives your audio its unique sound.\n1. Cooking with a Prompt Speech (Following a Famous Recipe)\n - A prompt speech provides the desired acoustic characteristics for VoxCPM. The speaker's timbre, speaking style, and even the background sounds and ambiance will be replicated.\n - For a Clean, Studio-Quality Voice:\n - \u2705 Enable \"Prompt Speech Enhancement\". This acts like a noise filter, removing background hiss and rumble to give you a pure, clean voice clone.\n2. Cooking au Naturel (Letting the Model Improvise)\n - If no reference is provided, VoxCPM becomes a creative chef! It will infer a fitting speaking style based on the text itself, thanks to the text-smartness of its foundation model, MiniCPM-4.\n - Pro Tip: Challenge VoxCPM with any text\u2014poetry, song lyrics, dramatic monologues\u2014it may deliver some interesting results!\n\n---\n### \ud83e\uddc2 Step 3: The Final Seasoning (Fine-Tuning Your Results)\nYou're ready to serve! But for master chefs who want to tweak the flavor, here are two key spices.\n- CFG Value (How Closely to Follow the Recipe)\n - Default: A great starting point.\n - Voice sounds strained or weird? Lower this value. It tells the model to be more relaxed and improvisational, great for expressive prompts.\n - Need maximum clarity and adherence to the text? Raise it slightly to keep the model on a tighter leash.\n- Inference Timesteps (Simmering Time: Quality vs. Speed)\n - Need a quick snack? Use a lower number. Perfect for fast drafts and experiments.\n - Cooking a gourmet meal? Use a higher number. This lets the model \"simmer\" longer, refining the audio for superior detail and naturalness.\n\n---\nHappy creating! \ud83c\udf89 Start with the default settings and tweak from there to suit your project. The kitchen is yours!\n\n\n---\n\n\n## \ud83c\udf1f Community Projects\n\nWe're excited to see the VoxCPM community growing! Here are some amazing projects and features built by our community:\n\n- **[ComfyUI-VoxCPM](https://github.com/wildminder/ComfyUI-VoxCPM)**\n- **[ComfyUI-VoxCPMTTS](https://github.com/1038lab/ComfyUI-VoxCPMTTS)** \n- **[WebUI-VoxCPM](https://github.com/rsxdalv/tts_webui_extension.vox_cpm)**\n- **[PR: Streaming API Support (by AbrahamSanders)](https://github.com/OpenBMB/VoxCPM/pull/26)** \n\n\n\n*Have you built something cool with VoxCPM? We'd love to feature it here! Please open an issue or pull request to add your project.*\n\n\n## \ud83d\udcca Performance Highlights\n\nVoxCPM achieves competitive results on public zero-shot TTS benchmarks:\n\n### Seed-TTS-eval Benchmark\n\n| Model | Parameters | Open-Source | test-EN | | test-ZH | | test-Hard | |\n|------|------|------|:------------:|:--:|:------------:|:--:|:-------------:|:--:|\n| | | | WER/%\u2b07 | SIM/%\u2b06| CER/%\u2b07| SIM/%\u2b06 | CER/%\u2b07 | SIM/%\u2b06 |\n| MegaTTS3 | 0.5B | \u274c | 2.79 | 77.1 | 1.52 | 79.0 | - | - |\n| DiTAR | 0.6B | \u274c | 1.69 | 73.5 | 1.02 | 75.3 | - | - |\n| CosyVoice3 | 0.5B | \u274c | 2.02 | 71.8 | 1.16 | 78.0 | 6.08 | 75.8 |\n| CosyVoice3 | 1.5B | \u274c | 2.22 | 72.0 | 1.12 | 78.1 | 5.83 | 75.8 |\n| Seed-TTS | - | \u274c | 2.25 | 76.2 | 1.12 | 79.6 | 7.59 | 77.6 |\n| MiniMax-Speech | - | \u274c | 1.65 | 69.2 | 0.83 | 78.3 | - | - |\n| CosyVoice | 0.3B | \u2705 | 4.29 | 60.9 | 3.63 | 72.3 | 11.75 | 70.9 |\n| CosyVoice2 | 0.5B | \u2705 | 3.09 | 65.9 | 1.38 | 75.7 | **6.83** | 72.4 |\n| F5-TTS | 0.3B | \u2705 | 2.00 | 67.0 | 1.53 | 76.0 | 8.67 | 71.3 |\n| SparkTTS | 0.5B | \u2705 | 3.14 | 57.3 | 1.54 | 66.0 | - | - |\n| FireRedTTS | 0.5B | \u2705 | 3.82 | 46.0 | 1.51 | 63.5 | 17.45 | 62.1 |\n| FireRedTTS-2 | 1.5B | \u2705 | 1.95 | 66.5 | 1.14 | 73.6 | - | - |\n| Qwen2.5-Omni | 7B | \u2705 | 2.72 | 63.2 | 1.70 | 75.2 | 7.97 | **74.7** |\n| OpenAudio-s1-mini | 0.5B | \u2705 | 1.94 | 55.0 | 1.18 | 68.5 | - | - |\n| IndexTTS2 | 1.5B | \u2705 | 2.23 | 70.6 | 1.03 | 76.5 | - | - |\n| VibeVoice | 1.5B | \u2705 | 3.04 | 68.9 | 1.16 | 74.4 | - | - |\n| HiggsAudio-v2 | 3B | \u2705 | 2.44 | 67.7 | 1.50 | 74.0 | - | - |\n| **VoxCPM** | 0.5B | \u2705 | **1.85** | **72.9** | **0.93** | **77.2** | 8.87 | 73.0 |\n\n\n### CV3-eval Benchmark\n\n| Model | zh | en | hard-zh | | | hard-en | | |\n|-------|:--:|:--:|:-------:|:--:|:--:|:-------:|:--:|:--:|\n| | CER/%\u2b07 | WER/%\u2b07 | CER/%\u2b07 | SIM/%\u2b06 | DNSMOS\u2b06 | WER/%\u2b07 | SIM/%\u2b06 | DNSMOS\u2b06 |\n| F5-TTS | 5.47 | 8.90 | - | - | - | - | - | - |\n| SparkTTS | 5.15 | 11.0 | - | - | - | - | - | - |\n| GPT-SoVits | 7.34 | 12.5 | - | - | - | - | - | - |\n| CosyVoice2 | 4.08 | 6.32 | 12.58 | 72.6 | 3.81 | 11.96 | 66.7 | 3.95 |\n| OpenAudio-s1-mini | 4.00 | 5.54 | 18.1 | 58.2 | 3.77 | 12.4 | 55.7 | 3.89 |\n| IndexTTS2 | 3.58 | 4.45 | 12.8 | 74.6 | 3.65 | - | - | - |\n| HiggsAudio-v2 | 9.54 | 7.89 | 41.0 | 60.2 | 3.39 | 10.3 | 61.8 | 3.68 |\n| CosyVoice3-0.5B | 3.89 | 5.24 | 14.15 | 78.6 | 3.75 | 9.04 | 75.9 | 3.92 |\n| CosyVoice3-1.5B | 3.91 | 4.99 | 9.77 | 78.5 | 3.79 | 10.55 | 76.1 | 3.95 |\n| **VoxCPM** | **3.40** | **4.04** | 12.9 | 66.1 | 3.59 | **7.89** | 64.3 | 3.74 |\n\n\n\n\n\n\n\n\n\n\n\n\n## \u26a0\ufe0f Risks and limitations\n- General Model Behavior: While VoxCPM has been trained on a large-scale dataset, it may still produce outputs that are unexpected, biased, or contain artifacts.\n- Potential for Misuse of Voice Cloning: VoxCPM's powerful zero-shot voice cloning capability can generate highly realistic synthetic speech. This technology could be misused for creating convincing deepfakes for purposes of impersonation, fraud, or spreading disinformation. Users of this model must not use it to create content that infringes upon the rights of individuals. It is strictly forbidden to use VoxCPM for any illegal or unethical purposes. We strongly recommend that any publicly shared content generated with this model be clearly marked as AI-generated.\n- Current Technical Limitations: Although generally stable, the model may occasionally exhibit instability, especially with very long or expressive inputs. Furthermore, the current version offers limited direct control over specific speech attributes like emotion or speaking style.\n- Bilingual Model: VoxCPM is trained primarily on Chinese and English data. Performance on other languages is not guaranteed and may result in unpredictable or low-quality audio.\n- This model is released for research and development purposes only. We do not recommend its use in production or commercial applications without rigorous testing and safety evaluations. Please use VoxCPM responsibly.\n\n\n\n## \ud83d\udcddTO-DO List\nPlease stay tuned for updates!\n- [x] Release the VoxCPM technical report.\n- [ ] Support higher sampling rate (next version).\n\n\n\n## \ud83d\udcc4 License\nThe VoxCPM model weights and code are open-sourced under the [Apache-2.0](LICENSE) license.\n\n## \ud83d\ude4f Acknowledgments\n\nWe extend our sincere gratitude to the following works and resources for their inspiration and contributions:\n\n- [DiTAR](https://arxiv.org/abs/2502.03930) for the diffusion autoregressive backbone used in speech generation\n- [MiniCPM-4](https://github.com/OpenBMB/MiniCPM) for serving as the language model foundation\n- [CosyVoice](https://github.com/FunAudioLLM/CosyVoice) for the implementation of Flow Matching-based LocDiT\n- [DAC](https://github.com/descriptinc/descript-audio-codec) for providing the Audio VAE backbone\n\n## Institutions\n\nThis project is developed by the following institutions:\n- <img src=\"assets/modelbest_logo.png\" width=\"28px\"> [ModelBest](https://modelbest.cn/)\n\n- <img src=\"assets/thuhcsi_logo.png\" width=\"28px\"> [THUHCSI](https://github.com/thuhcsi)\n\n\n## \u2b50 Star History\n [](https://star-history.com/#OpenBMB/VoxCPM&Date)\n\n\n## \ud83d\udcda Citation\n\nIf you find our model helpful, please consider citing our projects \ud83d\udcdd and staring us \u2b50\ufe0f\uff01\n\n```bib\n@article{voxcpm2025,\n title = {VoxCPM: Tokenizer-Free TTS for Context-Aware Speech Generation and True-to-Life Voice Cloning},\n author = {Zhou, Yixuan and Zeng, Guoyang and Liu, Xin and Li, Xiang and Yu, Renjie and Wang, Ziyang and Ye, Runchuan and Sun, Weiyue and Gui, Jiancheng and Li, Kehan and Wu, Zhiyong and Liu, Zhiyuan},\n journal = {arXiv preprint arXiv:2509.24650},\n year = {2025},\n}\n```\n",
"bugtrack_url": null,
"license": null,
"summary": "VoxCPM: Tokenizer-Free TTS for Context-Aware Speech Generation and True-to-Life Voice Cloning",
"version": "1.0.5",
"project_urls": {
"Bug Tracker": "https://github.com/OpenBMB/VoxCPM/issues",
"Documentation": "https://github.com/OpenBMB/VoxCPM#readme",
"Homepage": "https://github.com/OpenBMB/VoxCPM",
"Repository": "https://github.com/OpenBMB/VoxCPM.git"
},
"split_keywords": [
"voxcpm",
" text-to-speech",
" tts",
" speech-synthesis",
" voice-cloning",
" ai",
" deep-learning",
" pytorch"
],
"urls": [
{
"comment_text": null,
"digests": {
"blake2b_256": "f53e423139fe326ff5370d5f86e47ebb596cb8666c9364741a84cb4bb4d05bae",
"md5": "2cefc5780188d7970617aebaf47ebc87",
"sha256": "d3b860a1d63f3506e388b059f78982eba010706805e4750281ed560d2296ba41"
},
"downloads": -1,
"filename": "voxcpm-1.0.5-py3-none-any.whl",
"has_sig": false,
"md5_digest": "2cefc5780188d7970617aebaf47ebc87",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.10",
"size": 42583,
"upload_time": "2025-10-09T05:23:30",
"upload_time_iso_8601": "2025-10-09T05:23:30.136465Z",
"url": "https://files.pythonhosted.org/packages/f5/3e/423139fe326ff5370d5f86e47ebb596cb8666c9364741a84cb4bb4d05bae/voxcpm-1.0.5-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": null,
"digests": {
"blake2b_256": "0e940fb57fb764d25a9dec0f5842b0799c19eaa88573780e4e84d94d59e3e031",
"md5": "5f8b0b378a2e032e42c577d727a651ee",
"sha256": "6b631a366e4a59a0a9a9ed3080ef1fd531aaa0b1bb19a6ae1ded5baae050fafa"
},
"downloads": -1,
"filename": "voxcpm-1.0.5.tar.gz",
"has_sig": false,
"md5_digest": "5f8b0b378a2e032e42c577d727a651ee",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.10",
"size": 1493588,
"upload_time": "2025-10-09T05:23:31",
"upload_time_iso_8601": "2025-10-09T05:23:31.716848Z",
"url": "https://files.pythonhosted.org/packages/0e/94/0fb57fb764d25a9dec0f5842b0799c19eaa88573780e4e84d94d59e3e031/voxcpm-1.0.5.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2025-10-09 05:23:31",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "OpenBMB",
"github_project": "VoxCPM",
"travis_ci": false,
"coveralls": false,
"github_actions": true,
"lcname": "voxcpm"
}
JSON
