Aug 22, 2024 18:50 Upload 2 files Newly generated dev and train datasets with more focus on the 40-60 percent overlapped speech. Aug 19, 2024 19:06 Upload 3 files Uploading newly generated datasets for dev and train with no augmentations and higher overlap conversation focus, test set with no augmentations Jul 29, 2024 21:51 Upload 2 files Newly generated conversational overlap focussed (train and dev) datasets with no augmentations. Jul 24, 2024 21:52 Upload 2 files Adding new train and dev splits with no augmentations and data split focussed on on conversational overlap, less focus on 0 overlap, more focus on 20-60 percent overlap and slightly more focus on 80-10 percent overlap. 16:58 Upload 1_or_2_mix_extract_no_augmentations_present_only.tar Adding the test set for present only scenario (with no augmentations) 16:51 Upload 2 files Added two new datasets one for dev and train with no augmentations and target speaker present scenario only Jul 15, 2024 18:24 Upload 2 files Newly generated train and dev sets with no augmentations (the previous versions incorrectly still contained augmentations) 15:28 Upload 2 files Train and dev set with no augmentations. 15:10 Delete data/dev-no-augmentations remove faulty named dev folder 15:10 Delete data/train-no-augmentations remove faulty named train folder 15:04 Upload 1_or_2_mix_extract_no_augmentations.tar Train set with no augmentations 14:37 Upload 1_or_2_mix_extract_no_augmentations.tar Dev set with no augmentations. Jul 12, 2024 17:26 Upload 3 files Newly generated dev, train and test datasets - train set is now much larger 14:35 Delete data/train Removing old version of train dataset 14:35 Delete data/test Removing old version of test dataset 14:35 Delete data/dev Removing old version of dev dataset Jul 9, 2024 12:57 Upload 3 files Re-generated test, dev and train datasets - now contains augmentation information in metadata and augmented audio clips. 12:38 Delete data/validate Remove the validation dataset - we will use the dev dataset as our validation set during training May 31, 2024 17:57 Upload 2_3_mix_extract.tar Updated test dataset with the updated speaker extraction scenario logic. 17:34 Upload 3 files Updated dev, train and validate dataset with new speaker extraction overlap logic. May 17, 2024 20:06 Upload 4 files Re-generated all datasets with latest update to dataset gen implementation. May 15, 2024 12:31 Upload 2_3_mix_extract.tar New test set with latest dataset format (always add a ref entry in the metadata). 01:34 Upload 3 files Another update with ref always included in meta data May 14, 2024 18:34 Upload 3 files Updated dev, train and validate datasets after fixing a bug where the target speaker clip contained a clip in the s1/ref speaker clip. May 10, 2024 14:12 Upload 2_3_mix_extract.tar Added the test dataset generated from sample of librispeech test-clean May 9, 2024 23:10 Upload 3 files Updated all datasets to include the ref wav file which is just silent audio for the target speaker absent scenario. 21:02 Upload 4 files Added the speaker extraction datasets for dev, training and validation. Updated the readme with the dataset config. 
