Dataset for Interspeech 2018 submission: Singing voice phoneme segmentation by hierarchically inferring syllable and phoneme onset positions

This dataset contains the materials for training, testing the joint and HSMM models mentioned in the paper "Singing voice phoneme segmentation by hierarchically inferring syllable and phoneme onset positions".

The filename list of this dataset can be found in the function get_train_test_recordings_joint() of ./general/trainTestSeparation.py file. The dataset contains the Praat TextGrids and .wavs of the variables: train_primary_school, val_primary_school and test_primary_school. For accessing other datasets such as train_nacta_2017, train_nacta and train_sepa, please download them from the links:

jingju dataset part1: https://zenodo.org/record/1185154

jingju dataset part2: https://doi.org/10.5281/zenodo.842229

Once you have downloaded these three datasets, you need to set the paths in ./general/filePathShared.py.

Set path_jingju_dataset to the parent path of these three datasets.

Set primarySchool_dataset_root_path to the path of the interspeech2018 dataset (the current dataset).

Set nacta_dataset_root_path to the path of the jingju dataset part1.

Set nacta2017_dataset_root_path to the path the jingju dataset part2.

For more information on this paper, please refer to the Github page: https://github.com/ronggong/interspeech2018_submission01