Dataset · espnet
ACE-KiSing singing voice segments
espnet/ace-kising-segments
Vocal audio recordings and lyrics in Chinese and English for singing generation and recognition.
What's inside
The dataset contains between 10k and 100k pairs of singing voice audio snippets and text in Chinese and English. It was presented as part of a 2024 research paper on data scaling for vocal synthesis (ACE-Opencpop and ACE-KiSing) and is suitable for text-to-audio, audio-to-audio, and ASR tasks.
What you can do with it
- Training text-to-singing voice synthesis (SVS) models
- Fine-tuning automatic speech recognition (ASR) systems on vocal tracks
- Tasks
- Text → audio, Audio → audio, Automatic speech recognition
- Languages
- Chinese, English
- Size
- 10K–100K rows
- Downloads this month
- 451
- Likes
- 8
How to load it
from datasets import load_dataset
ds = load_dataset("espnet/ace-kising-segments")