Dataset Introduction
MagicData-Dialect-Cantonese-TTS-Lite is an open-source Cantonese TTS subset of the MagicData-Dialect-TTS-Lite collection released by Magic Data. It focuses on authentic Cantonese speech from Guangzhou and is designed for research scenarios such as dialect speech synthesis, acoustic analysis, and model evaluation.
The dataset contains approximately 10 minutes of speech data, recorded by one native Cantonese speaker from Guangzhou. The speaker was born and raised in the local region, and the recordings preserve authentic local accent, intonation, and expression habits.
Dataset Overview
| Dialect | City | Code | Duration | Sentences | Speaker |
|---|---|---|---|---|---|
| Cantonese | Guangzhou | GUD | 10 minutes | 54 sentences | 1 female, 55 years old |
Dataset Features
1. Native speaker with authentic accent
- The speaker was born and raised in the local region until adulthood.
- The speaker’s family and main social environment use the local dialect.
2. Daily-life content coverage
- Weather, food, family conversations, numbers, time, and dates
- A small amount of emotional expression
- No complex technical terms, news reading, or poetry recitation, in order to avoid style deviation
3. Clean recording environment
- Quiet indoor environment
- 48 kHz / 16-bit / mono WAV
4. Moderate sentence length, suitable for TTS modeling
- Each sentence is around 5–20 seconds, with an average length of about 10 seconds
- Natural punctuation-based segmentation, with no forced truncation
Annotation Guidelines
- Chinese character transcription: Standard Chinese characters are used, while dialect-specific words are preserved and restored.
- Number annotation: Numbers are written in Chinese character form.
- Standardization rule: The original dialect sentences are preserved and are not forcibly “translated” into Mandarin.
Open-source File Structure
dialect-tts-lite/
└──广东
│ ├── ProsodyLabeling
│ │ ├── txt
│ ├── wav
│ │ ├── wav/ # 54个音频文件
│ └── 2026自研短语音转写规范——中文
Usage Recommendations
Suitable for:
- Zero-shot / few-shot baseline testing for multi-dialect TTS models
- Acoustic analysis of dialectal phonetic features
- Comparative experiments in academic research
Not suitable for:
- Directly training production-level dialect TTS products, as the dataset is non-commercial and limited in scale
- Evaluating extreme scenarios, such as noisy environments, far-field recording, or children’s voices
If you are interested in a larger-scale commercial version, please contact us.
Open-source License
This dataset is for non-commercial use only under the CC BY-NC-ND 4.0 license. It is suitable for academic research, personal development, and model evaluation.
📧 For the full commercial version, please contact: business@magicdatatech.com
