Dataset · AISHELL
Multilingual Mandarin speech corpus for TTS
AISHELL/AISHELL-3
A large neutral Mandarin speech corpus from 218 speakers with gender, age, and accent annotations for TTS training.
What's inside
The dataset contains about 85 hours of high-quality Mandarin audio recordings divided into 88,035 utterances. The speech is recorded in a neutral tone by 218 native speakers, with gender, age group, and native accent specified for each, along with corresponding transcriptions. The corpus was published by Beijing Shell Shell Technology.
What you can do with it
- Training multilingual Mandarin text-to-speech (TTS) systems
- Configuring voice cloning models and pitch control via speaker metadata
- Training automated speech recognition (ASR) and accent classification systems
- Tasks
- Text → speech
- Languages
- Chinese
- Size
- 10K–100K rows
- Downloads this month
- 62.6K
- Likes
- 17
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("AISHELL/AISHELL-3")