TAPS Dataset

Throat and Acoustic Pairing Speech Dataset for Deep Learning-Based Speech Enhancement

Large Scale Dataset

12.7 hours of training data with 5,000 paired utterances and 2.6 hours of evaluation data

Diverse Speakers

60 native Korean speakers with balanced gender distribution

High Quality

Synchronized recordings with professional equipment and careful post-processing

Latest Updates

January 2024

Dataset Release

Initial release of TAPS dataset with baseline models

December 2023

Paper Publication

Research paper accepted and published