Dataset · MrSupW
Contextual automatic speech recognition (ASR) benchmark
MrSupW/ContextASR-Bench
Massive audio and text corpus in Chinese and English for testing linguistic knowledge of ASR models.
What's inside
The ContextASR-Bench benchmark includes 10k to 100k records in audio and text modalities in Chinese and English. Unlike standard acoustic robustness tests, this set evaluates linguistic capabilities and background knowledge of speech recognition systems.
What you can do with it
- Evaluating contextual speech recognition quality and complex terminology in ASR models
- Comparing modern speech models on context understanding levels in English and Chinese
- Tasks
- Automatic speech recognition
- Languages
- Chinese, English
- Size
- 10K–100K rows
- Downloads this month
- 2.6K
- Likes
- 38
How to load it
from datasets import load_dataset
ds = load_dataset("MrSupW/ContextASR-Bench")