CS-Dialogue

# CS-Dialogue: A 104-Hour Dataset of Spontaneous Mandarin-English Code-Switching Dialogues for Speech Recognition

[](https://huggingface.co/datasets/BAAI/CS-Dialogue)
[](https://arxiv.org/abs/2502.18913)
[](https://creativecommons.org/licenses/by-nc-sa/4.0/)

## Introduction

**CS-Dialogue** is a large-scale, publicly available Mandarin-English code-switching speech dialogue dataset. This dataset solves key problems found in existing code-switching speech datasets — mainly their small size, lack of natural conversations, and missing full-length dialogue recordings. It provides a solid foundation for advancing research in code-switching ASR and other related fields. The dataset is released under a **CC BY-NC-SA 4.0 license**, meaning it is available for non-commercial use.

## Dataset Details

This dataset contains 104.02 hours of spontaneous dialogue recordings, consisting of 100 pairs of two-person conversations recorded by 200 speakers. Key features of the dataset include:

* **Speakers:** 200 speakers with strong English proficiency (e.g., IELTS ≥ 6 or passing TEM-4).
* **Geographic Diversity:** Speakers come from 30 provincial-level regions across mainland China.
* **Content:** Each conversation covers 2 to 6 topics and includes Mandarin-only, code-switching, and English-only segments.
* **Audio Format:** WAV files with a 16kHz sampling rate.
* **Transcriptions:** Carefully crafted, character-level manual transcriptions.
* **Annotations:** The dataset includes annotations for each utterance, and for the speakers level.
* **Utterance-level**: `id`, `audio` (file path), `text` (transcription).
* **Speaker-level**: `speaker_id`, `age`, `gender`, `location` (province), `device`.

### Dataset Structure

The dataset file structure is as follows.

```
data
├── long_wav/*.tar.gz
├── short_wav/*.tar.gz
└── index
├── long_wav
│ ├── dev.txt
│ ├── test.txt
│ └── train.txt
├── short_wav
│ ├── dev
│ │ ├── text
│ │ └── wav.scp
│ ├── test
│ │ ├── text
│ │ └── wav.scp
│ └── train
│ ├── text
│ └── wav.scp
└── total_infomation
└── Information_Index.txt
```

For more details, please refer to our paper [CS-Dialogue](https://arxiv.org/abs/2502.18913).

## 📚 Cite me

```
@article{zhou2025cs,
title={CS-Dialogue: A 104-Hour Dataset of Spontaneous Mandarin-English Code-Switching Dialogues for Speech Recognition},
author={Zhou, Jiaming and Guo, Yujie and Zhao, Shiwan and Sun, Haoqin and Wang, Hui and He, Jiabei and Kong, Aobo and Wang, Shiyao and Yang, Xi and Wang, Yequan and others},
journal={arXiv preprint arXiv:2502.18913},
year={2025}
}
```

# CS-Dialogue：面向语音识别的104小时自然普通话-英语语码转换对话数据集

[](https://huggingface.co/datasets/BAAI/CS-Dialogue)
[](https://arxiv.org/abs/2502.18913)
[](https://creativecommons.org/licenses/by-nc-sa/4.0/)

## 简介

**CS-Dialogue** 是大规模、可公开获取的普通话-英语语码转换语音对话数据集。本数据集解决了现有语码转换语音数据集存在的核心痛点：体量偏小、缺乏自然对话场景、缺少完整时长的对话录音，为推动语码转换自动语音识别（Automatic Speech Recognition, ASR）及相关领域的研究提供了坚实基础。该数据集采用**CC BY-NC-SA 4.0协议**发布，允许非商业性使用。

## 数据集详情

本数据集包含104.02小时的自然对话录音，由200名说话者录制的100组双人对话组成。该数据集的核心特性如下：
* **说话者信息**：共200名英语能力出众的说话者（例如雅思成绩≥6分或通过英语专业四级考试）。
* **地域多样性**：说话者来自中国大陆30个省级行政区。
* **对话内容**：每组对话涵盖2至6个主题，包含纯普通话、语码转换及纯英语片段。
* **音频格式**：采样率为16kHz的WAV格式文件。
* **转写文本**：经过精心打磨的字符级人工转写结果。
* **标注信息**：数据集包含话语级与说话者级别的两类标注：
* **话语级标注**：`id`、`audio`（文件路径）、`text`（转写文本）。
* **说话者级标注**：`speaker_id`、`age`（年龄）、`gender`（性别）、`location`（所属省份）、`device`（录制设备）。

### 数据集文件结构

数据集的文件目录结构如下：

data
├── long_wav/*.tar.gz
├── short_wav/*.tar.gz
└── index
├── long_wav
│ ├── dev.txt
│ ├── test.txt
│ └── train.txt
├── short_wav
│ ├── dev
│ │ ├── text
│ │ └── wav.scp
│ ├── test
│ │ ├── text
│ │ └── wav.scp
│ └── train
│ ├── text
│ └── wav.scp
└── total_infomation
└── Information_Index.txt

如需获取更多细节，请参阅我们的论文[CS-Dialogue](https://arxiv.org/abs/2502.18913)。

## 📚 引用格式

@article{zhou2025cs,
title={CS-Dialogue: A 104-Hour Dataset of Spontaneous Mandarin-English Code-Switching Dialogues for Speech Recognition},
author={Zhou, Jiaming and Guo, Yujie and Zhao, Shiwan and Sun, Haoqin and Wang, Hui and He, Jiabei and Kong, Aobo and Wang, Shiyao and Yang, Xi and Wang, Yequan and others},
journal={arXiv preprint arXiv:2502.18913},
year={2025}
}