Dataset · ASLP-lab
Multilingual multi-speaker conversational ASR benchmark
ASLP-lab/MS-Bench
32.63 hours of multi-voice recordings in 6 languages with speech overlap for testing recognition systems.
What's inside
The MS-Bench benchmark includes 32.63 hours of meeting and dialogue audio recordings ranging from 1.38 to 159 minutes across six languages (Chinese, English, Portuguese, Japanese, Thai, and Italian). Each recording features 2 to 14 speakers, with an average speech overlap of 21.87%.
What you can do with it
- Evaluating ASR system quality in complex multi-speaker and overlapping speech conditions
- Comparing multilingual long-meeting transcription models
- Tasks
- Automatic speech recognition
- Languages
- Chinese, English, Portuguese, Japanese, Thai, Italian, Spanish
- Size
- up to 1K rows
- Downloads this month
- 772
- Likes
- 3
How to load it
from datasets import load_dataset
ds = load_dataset("ASLP-lab/MS-Bench")