acidcoma/ru_librispeech_for_speaker_separation

Dataset for source audio separation task based on Russian LibriSpeech (RuLS) dataset. Dataset contains 50 000 audio mixtures with 2 speakers for train part; 12500 audio mixtures for test part. Dataset also containts metadata files with audio duration (sec), source 1 and source 2 filepaths for each audio mixture. source: https://www.openslr.org/96/