Skip to content

Dataset · amphion

Chinese speech with non-verbal sound annotations

amphion/Emilia-NV

Mandarin Chinese audio and text with word-level annotations of sighs, laughter, and interjections for ASR and TTS.

What's inside

Contains Mandarin Chinese audio recordings and texts (100k to 1M examples). The key feature is word-level annotation of 18 paralinguistic sound categories, including laughter, breathing, and interjections like uhm or oh. The data is designed to improve speech synthesis and recognition systems.

What you can do with it

  • Training emotional and expressive text-to-speech (TTS) models in Chinese
  • Fine-tuning ASR systems to recognize non-verbal sounds and speech pauses
  • Evaluating the quality of natural intonation generation in voice models
Tasks
Text → speech, Automatic speech recognition
Languages
Chinese
Size
100K–1M rows
Downloads this month
625
Likes
56
Access
on request on Hugging Face

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("amphion/Emilia-NV")