tencent/AuK · Hugging Face

来源：https://huggingface.co/tencent/AuK

https://huggingface.co/tencent/AuK#news动态

• [2026/09/09] 🎉 AuK现已开源。代码和模型权重均已公开发布。欢迎在🤗演示空间（https://huggingface.co/spaces/tencent/AuK）或🤖ModelScope空间（https://modelscope.cn/studios/Tencent-Hunyuan/AuK）进行试用！

https://huggingface.co/tencent/AuK#contents目录

• 新动态 (https://huggingface.co/tencent/AuK#news)
• 简介 (https://huggingface.co/tencent/AuK#introduction)
• 性能表现 (https://huggingface.co/tencent/AuK#performance)
• 模型架构 (https://huggingface.co/tencent/AuK#model-architecture)
• 支持的任务 (https://huggingface.co/tencent/AuK#supported-tasks)
• 下载权重 (https://huggingface.co/tencent/AuK#download-the-weights)
• 引用方式 (https://huggingface.co/tencent/AuK#citation)
• 开源协议 (https://huggingface.co/tencent/AuK#license)

https://huggingface.co/tencent/AuK#introduction简介

AuK是一个用于语音生成与编辑的15亿参数基础模型。它在数百万小时多样音频数据上训练，通过统一的自然语言指令接口，支持零样本及基于指令的语音合成、内容与声学编辑、副语言编辑、语音增强和源分离。AuK包含两个变体：

本仓库包含AuK基础模型的官方权重。

https://huggingface.co/tencent/AuK#performance性能表现

AuK在语音生成、编辑、增强和分离基准测试中的表现（https://huggingface.co/tencent/AuK/blob/main/assets/performance.png）

https://huggingface.co/tencent/AuK#model-architecture模型架构

模型架构（https://huggingface.co/tencent/AuK/blob/main/assets/arch.png）

https://huggingface.co/tencent/AuK#supported-tasks支持的任务

AuK通过统一的自然语言指令接口提供所有任务。下表按类别对支持的任务进行了分组，并附有简短描述及对应操作指南（Cookbook）章节链接，其中包含指令模板以及CLI和Python代码示例。

https://huggingface.co/tencent/AuK#download-the-weights下载权重

🤗 HuggingFace

pip install -U "huggingface_hub[cli]"

# AuK-Base
hf download tencent/AuK --local-dir ./ckpts/AuK

# AuK-Flash (4步蒸馏版)
hf download tencent/AuK-Flash --local-dir ./ckpts/AuK-Flash

# 多模态大语言模型编码器
hf download Qwen/Qwen2.5-Omni-3B  --local-dir ./ckpts/Qwen2.5-Omni-3B

🤖 ModelScope

pip install -U modelscope

# AuK-Base
modelscope download --model Tencent-Hunyuan/AuK --local_dir ./ckpts/AuK

# AuK-Flash (4步蒸馏版)
modelscope download --model Tencent-Hunyuan/AuK-Flash  --local_dir ./ckpts/AuK-Flash

# 多模态大语言模型编码器
modelscope download --model Qwen/Qwen2.5-Omni-3B --local_dir ./ckpts/Qwen2.5-Omni-3B

预期的目录结构如下：

ckpts/
├── AuK/
├── AuK-Flash/          # 可选
└── Qwen2.5-Omni-3B/

模型检查点包含扩散Transformer和层融合权重。多模态大语言模型编码器和VAE在运行时从单独的文件加载，因此在加载检查点时缺少 text_encoder.* 键是正常现象。

关于安装、推理、Gradio、ComfyUI和微调，请参阅GitHub README（https://github.com/Tencent-Hunyuan/AuK#quick-start）和操作指南（Cookbook）（https://github.com/Tencent-Hunyuan/AuK/blob/main/docs/COOKBOOK.md）。

https://huggingface.co/tencent/AuK#inference-with-sglang-omni使用SGLang-Omni进行推理

感谢SGLang Omni团队在发布当天即完成对AuK的支持。你可以使用以下命令启动SGLang Omni：

python -m sglang_omni.cli serve --model-path tencent/AuK

有关详细用法和配置选项，请参阅SGLang Omni AuK操作指南（https://sgl-project.github.io/sglang-omni/cookbook/auk.html）和AuK的优化路线图（https://github.com/sgl-project/sglang-omni/issues/2064）。

https://huggingface.co/tencent/AuK#citation引用方式

使用AuK进行研究时，可引用以下条目：

@misc{ma2026auktechnicalreportopensource,
  title={AuK Technical Report: An Open-Source Foundational Model for Speech Generation and Editing},
  author={Ziyang Ma and Zhikang Niu and Wenming Tu and Tianrui Wang and Ruiqi Yan and Junxi Liu and Yanru Huo and Nickk Huang and Yang Liu and Qicong Xie and Zeyu Xie and Hui Wang and Haitao Li and Zixuan Jiang and Yalin Li and Jie Fang and Yifan Duan and Zeyue Tian and Guangzheng Li and Haina Zhu and Shuyi Wang and Jinwen Wang and Mingyu Cui and Tian Tan and Auden and Sen Liang and Steve Yves and Shan Yang and Liefeng Bo and Zilong Zheng and Kai Yu and Eng-Siong Chng and Xie Chen},
  year={2026},
  eprint={2609.08936},
  archivePrefix={arXiv},
  primaryClass={cs.SD},
  url={https://arxiv.org/abs/2609.08936},
}

https://huggingface.co/tencent/AuK#license开源协议

AuK基于MIT协议发布。完整条款请参见LICENSE文件（https://huggingface.co/tencent/AuK/tree/main/LICENSE）。