ESC-50
A public benchmark dataset for environmental sound classification, containing 2000 audio clips across 50 classes. It is commonly used in research on audio event recognition.
Timeline (last 90 days)
Of the non-drone clips in the DroneAudioDataset, 96.4% originate from the ESC-50 sound dataset.
The dual-view stacking model correctly classifies 98.78% of unseen ESC-50 source clips as non-drone.
In a sound classification experiment on the ESC-50 dataset, the top-1 hit rate was approximately 25% before training and rose to approximately 66% after about seven and a half minutes of training on an A100.
In a test of 50 ESC-50 everyday sound clips, the correct clip ranked first only 24% of the time (versus 2% for random guessing), revealing that ambient sound effects are a weakness of the model.
All Facts (4)
Of the non-drone clips in the DroneAudioDataset, 96.4% originate from the ESC-50 sound dataset.
50%UnverifiedThe dual-view stacking model correctly classifies 98.78% of unseen ESC-50 source clips as non-drone.
50%UnverifiedIn a sound classification experiment on the ESC-50 dataset, the top-1 hit rate was approximately 25% before training and rose to approximately 66% after about seven and a half minutes of training on an A100.
50%UnverifiedIn a test of 50 ESC-50 everyday sound clips, the correct clip ranked first only 24% of the time (versus 2% for random guessing), revealing that ambient sound effects are a weakness of the model.
50%