Dataset · amphion
Emilia multilingual speech corpus
amphion/Emilia-Dataset
Multilingual audio and text for speech generation: 101k hours original and over 200k in Emilia-Large.
What's inside
The official repository of the Emilia dataset for large-scale speech generation. Includes audio and text in Chinese, English, Japanese, French, German, and Korean. Emilia-Large combines the original 101k hours (CC BY-NC 4.0) with an additional 114k hours, totaling over 200k hours. The Emilia-Pipe code for speech preprocessing has also been published.
What you can do with it
- Training TTS models
- Fine-tuning ASR on multilingual speech
- Creating voice models and evaluating synthesis
- Tasks
- Text → speech, Automatic speech recognition
- Languages
- Chinese, English, Japanese, French, German, Korean
- Size
- 10M–100M rows
- Downloads this month
- 89.7K
- Likes
- 520
- Access
- on request on Hugging Face
Models trained on it
- remsky/kokoro-inno-clone-tunerText → speech 1.7M
- SWivid/F5-TTSText → speech 1.3M
- SWivid/E2-TTSText → speech 71.4K
- neuphonic/neucodecAudio → audio 55K
- nvidia/magpie_tts_multilingual_357mText → speech 16.5K
- Aratako/MioCodec-25Hz-24kHzAudio → audio 9K
- neuphonic/neutts-air-q4-ggufText → speech 5.2K
- marduk-ra/F5-TTS-GermanText → speech 2.9K
How to load it
from datasets import load_dataset
ds = load_dataset("amphion/Emilia-Dataset")