Dataset · Hothan
Real-time duplex multimodal interaction benchmark
Hothan/Omni-DuplexEval
Evaluation dataset for streaming audio-text models that decide what and when to respond.
What's inside
A compact benchmark (fewer than 1,000 records) containing audio and text data to test multimodal models in real-time streaming mode. Includes two scenarios: Real-Time Description and Proactive Reminder.
What you can do with it
- Evaluating streaming voice and multimodal assistants
- Testing models' ability to determine the moment to join the dialogue (turn-taking)
- Size
- up to 1K rows
- Downloads this month
- 43.5K
- Likes
- 4
How to load it
from datasets import load_dataset
ds = load_dataset("Hothan/Omni-DuplexEval")