Dataset · nonverbalspeech
Audio dataset of non-verbal speech from movies and audiobooks
nonverbalspeech/nonverbalspeech38k
38,718 audio clips in English and Chinese with non-verbal sound annotations for TTS and ASR tasks.
What's inside
The dataset contains 38,718 text-aligned audio recordings with a total duration of about 131 hours in English and Chinese. Samples are collected from real-world sources like movies, cartoons, and audiobooks, and annotated across 10 categories of non-verbal speech manifestations.
What you can do with it
- Training TTS models to generate expressive speech with non-verbal sounds
- Fine-tuning ASR systems to recognize emotions and non-verbal insertions in audio
- Tasks
- Text → speech, Automatic speech recognition
- Languages
- Chinese, English
- Size
- 10K–100K rows
- Downloads this month
- 1.3K
- Likes
- 40
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("nonverbalspeech/nonverbalspeech38k")