mandipgoswami/

LibriRIR-100

	LibriRIR-100




	
	Dataset Summary

LibriRIR-100 is a large-scale paired clean↔reverberant speech training corpus containing exactly 100 hours of speech. Each utterance is paired with a room impulse response from RIR-Mega (mandipgoswami/rirmega), stratified across four RT60 reverberation conditions. Designed as a drop-in training resource for robust ASR, speech enhancement, and dereverberation models.

Existing paired reverberant speech datasets are… See the full description on the dataset page: https://huggingface.co/datasets/mandipgoswami/LibriRIR-100.