We present Kimi-Audio, an open-source audio foundation model excelling in **audio understanding, generation, and conversation**. This repository contains the official implementation, models, and evaluation toolkit for Kimi-Audio.
### 1、使用源码编译方式安装
## 🔥🔥🔥 News!!
#### 编译环境准备
* April 27, 2025: 👋 We release pretrained model weights of [Kimi-Audio-7B](https://huggingface.co/moonshotai/Kimi-Audio-7B).
提供基于fastpt不转码编译：
* April 25, 2025: 👋 We release the inference code and model weights of [Kimi-Audio-7B-Instruct](https://huggingface.co/moonshotai/Kimi-Audio-7B-Instruct).
* April 25, 2025: 👋 We release the audio evaluation toolkit [Kimi-Audio-Evalkit](https://github.com/MoonshotAI/Kimi-Audio-Evalkit). We can easily reproduce the **our results and baselines** by this toolkit!
* April 25, 2025: 👋 We release the technical report of [Kimi-Audio](https://arxiv.org/pdf/2504.18425).
Kimi-Audio is designed as a universal audio foundation model capable of handling a wide variety of audio processing tasks within a single unified framework. Key features include:
***Universal Capabilities:** Handle diverse tasks like automatic speech recognition (ASR), audio question answering (AQA), automatic audio captioning (AAC), speech emotion recognition (SER), sound event/scene classification (SEC/ASC), and end-to-end speech conversation.
***State-of-the-Art Performance:** Achieve SOTA results on numerous audio benchmarks (see [Evaluation](#evaluation) and the [Technical Report](https://arxiv.org/pdf/2504.18425)).
***Large-Scale Pre-training:** Pre-train on over 13 million hours of diverse audio data (speech, music, sounds) and text data, enabling robust audio reasoning and language understanding.
***Novel Architecture:** Employ a hybrid audio input (continuous acoustic vectors + discrete semantic tokens) and an LLM core with parallel heads for text and audio token generation.
***Efficient Inference:** Feature a chunk-wise streaming detokenizer based on flow matching for low-latency audio generation.
***Open-Source:** Release the code and model checkpoints for both pre-training and instruction fine-tuning, and release a comprehensive evaluation toolkit to foster community research and development.
## Architecture Overview
<palign="center">
<imgsrc="assets/kimia_framework.png"width="70%"/>
<p>
Kimi-Audio consists of three main components:
1.**Audio Tokenizer:** Converts input audio into:
* Discrete semantic tokens (12.5Hz) using vector quantization.
* Continuous acoustic features derived from a Whisper encoder (downsampled to 12.5Hz).
2.**Audio LLM:** A transformer-based model (initialized from a pre-trained text LLM like Qwen 2.5 7B) with shared layers processing multimodal inputs, followed by parallel heads for autoregressively generating text tokens and discrete audio semantic tokens.
3.**Audio Detokenizer:** Converts the predicted discrete semantic audio tokens back into high-fidelity waveforms using a flow-matching model and a vocoder (BigVGAN), supporting chunk-wise streaming with a look-ahead mechanism for low latency.
Here are performances on different benchmarks, you can easily reproduce the **our results and baselines** by our [Kimi-Audio-Evalkit](https://github.com/MoonshotAI/Kimi-Audio-Evalkit)(also see [**Evaluation Toolkit**](#evaluation-toolkit)):
<caption>Performance of Kimi-Audio and baseline models on speech conversation.</caption>
<thead>
<tr>
<throwspan="2">Model</th>
<thcolspan="6">Ability</th>
</tr>
<tr>
<th>Speed Control</th>
<th>Accent Control</th>
<th>Emotion Control</th>
<th>Empathy</th>
<th>Style Control</th>
<th>Avg</th>
</tr>
</thead>
<tbody>
<tr>
<td>GPT-4o</td>
<td>4.21</td>
<td><strong>3.65</strong></td>
<td>4.05</td>
<td><strong>3.87</strong></td>
<td><strong>4.54</strong></td>
<td><strong>4.06</strong></td>
</tr>
<tr>
<td>Step-Audio-chat</td>
<td>3.25</td>
<td>2.87</td>
<td>3.33</td>
<td>3.05</td>
<td>4.14</td>
<td>3.33</td>
</tr>
<tr>
<td>GLM-4-Voice</td>
<td>3.83</td>
<td>3.51</td>
<td>3.77</td>
<td>3.07</td>
<td>4.04</td>
<td>3.65</td>
</tr>
<tr>
<td>GPT-4o-mini</td>
<td>3.15</td>
<td>2.71</td>
<td>4.24</td>
<td>3.16</td>
<td>4.01</td>
<td>3.45</td>
</tr>
<tr>
<td>Kimi-Audio</td>
<td><strong>4.30</strong></td>
<td>3.45</td>
<td><strong>4.27</strong></td>
<td>3.39</td>
<td>4.09</td>
<td>3.90</td>
</tr>
</tbody>
</table>
## Evaluation Toolkit
Evaluating and comparing audio foundation models is challenging due to inconsistent metrics, varying inference configurations, and a lack of standardized generation evaluation. To address this, we developed and open-sourced an **Evaluation Toolkit**.
Key features:
* Integrates Kimi-Audio and other recent audio LLMs.
* Implements standardized metric calculation and integrates LLMs for intelligent judging (e.g., for AQA).
* Provides a unified platform for side-by-side comparisons with shareable inference 'recipes' for reproducibility.
* Includes a benchmark for evaluating speech conversation abilities (control, empathy, style).
We encourage the community to use and contribute to this toolkit to foster more reliable and comparable benchmarking. Find it here: [Kimi-Audio-Evalkit](https://github.com/MoonshotAI/Kimi-Audio-Evalkit).
## Generation Testset
We collect and release [Kimi-Audio-Generation-Testset](https://huggingface.co/datasets/moonshotai/Kimi-Audio-GenTest), which is designed to benchmark and evaluate the conversational capabilities of audio-based dialogue models. It consists of a collection of audio files containing various instructions and conversational prompts. The primary goal is to assess a model's ability to generate not just relevant, but also appropriately styled audio responses. The language in dataset is Chinese.
## License
The model is based and modified from [Qwen 2.5-7B](https://github.com/QwenLM/Qwen2.5). Code derived from Qwen2.5-7B is licensed under the [Apache 2.0 License](https://www.apache.org/licenses/LICENSE-2.0). Other parts of the code are licensed under the [MIT License](https://opensource.org/licenses/MIT).
## Acknowledgements
We would like to thank the following projects and individuals for their contributions to the development of Kimi-Audio:
Thank you to all the open-source projects for their contributions to this project!
## Citation
If you find Kimi-Audio useful in your research or applications, please cite our technical report:
```bibtex
## 验证
@misc{kimiteam2025kimiaudiotechnicalreport,
```
title={Kimi-Audio Technical Report},
python3
author={KimiTeam and Ding Ding and Zeqian Ju and Yichong Leng and Songxiang Liu and Tong Liu and Zeyu Shang and Kai Shen and Wei Song and Xu Tan and Heyi Tang and Zhengtao Wang and Chu Wei and Yifei Xin and Xinran Xu and Jianwei Yu and Yutao Zhang and Xinyu Zhou and Y. Charles and Jun Chen and Yanru Chen and Yulun Du and Weiran He and Zhenxing Hu and Guokun Lai and Qingcheng Li and Yangyang Liu and Weidong Sun and Jianzhou Wang and Yuzhi Wang and Yuefeng Wu and Yuxin Wu and Dongchao Yang and Hao Yang and Ying Yang and Zhilin Yang and Aoxiong Yin and Ruibin Yuan and Yutong Zhang and Zaida Zhou},
Python 3.10.12 (main, Feb 4 2025, 14:57:36) [GCC 11.4.0] on linux
year={2025},
Type "help", "copyright", "credits" or "license" for more information.
eprint={2504.18425},
>>> import kimia_infer
archivePrefix={arXiv},
>>> kimia_infer.__version__
primaryClass={eess.AS},
'0.1.0'
url={https://arxiv.org/abs/2504.18425},
>>>
}
```
```
版本号与官方版本同步，查询该软件的版本号，例如0.1.0；
## Contact Us
## Known Issue
- 无
For questions, issues, or collaboration inquiries, please feel free to open an issue on GitHub.
