About TAU

Learn about the design principles, methodology, and data behind the Taiwan Audio Understanding Benchmark

Design Principles and Data Flow

TAU follows a rigorous four-stage methodology to ensure high-quality, culturally-relevant audio understanding evaluation:

TAU Data Pipeline

[PLACEHOLDER: Add description of the pipeline methodology from paper]

Data Statistics Summary

Audio Length Distribution

Audio Length Distribution Chart

Median Length: 9.43 seconds

Range: 0 - 30 seconds

Question Type Distribution

Question Type Distribution Chart

Single-hop: 58.2% (1045 questions)

Multi-hop: 41.8% (749 questions)

Total Questions: 1794

Contact the Authors

Primary Author
Yi-Cheng Lin
even.dlion8@gmail.com
Graduate Institute of Communication Engineering, National Taiwan University
Supervisor
Hung-Yi Lee
hungyilee@ntu.edu.tw
Department of Electrical Engineering, National Taiwan University

For questions about the benchmark, dataset access, or research collaboration, please reach out to either author. We welcome feedback and are happy to discuss potential applications of the TAU benchmark.