An easy access of useful datasets for the DESED, domestic environment sound event detection dataset, including the weakly labelled real, strongly labelled real, strongly labelled external real (from StrongAudioSet) and DCASE development set. All datasets are provided with both wavforms and labels.
DESED dataset
The DESED dataset is used for SED research and DCASE challenge.
-
Weakly labeled data: 44.1kHz, multi-channel, weak data
-
Strongly labeled REAL data: 44.1kHz, multi-channel, strong data
-
Development data: 16kHz, multi-channel, development set
-
Extra strongly-labeled data extracted from the Strongly-labelled AudioSet: 16kHz, single channel, DESED_from_StrongAS
The last set is used in ATST-SED.