Releases: hearbenchmark/hear-preprocess
Releases · hearbenchmark/hear-preprocess
v2021.1.1
This version is a minor update containing changes to URLs and language following the renaming of GitHub organization from neuralaudio to hearbenchmark
What's Changed
Full Changelog: v2021.1.0...v2021.1.1
v2021.1.0
Version used for HEAR 2021 NeurIPS Evaluations
This version:
- Adds all the secret task config files from HEAR 2021
- Adds support for TensorFlow datasets
- Adds split_mode and max_task_duration_by_split in task_config
- Now supports k-fold splits in addition to train/val/test splits
- Adds support for rar files
- Validation for config files
v2021.0.5
- Open dataset preprocessing, as used in the leaderboard
- Substantial code cleanup and re-factoring
v2021.0.1
- Migrated preprocessing code from hear-eval-kit
- office_sounds is based audio and labels DCASE2016 Task 2, but with a completely different train/val/test split.
- speech_commands includes the full test set.