WHAMR!: Noisy and Reverberant Single-Channel Speech Separation
WHAMR!: Noisy and Reverberant Single-Channel Speech Separation
While significant advances have been made with respect to the separation of overlapping speech signals, studies have been largely constrained to mixtures of clean, near anechoic speech, not representative of many real-world scenarios. Although the WHAM! dataset introduced noise to the ubiquitous wsj0-2mix dataset, it did not include reverberation, which …