Dynamic Convolutional Neural Networks as Efficient Pre-trained Audio Models

Benchmark Model Rank Results
audio-classification-on-audiosetDyMN-L (Audio-Only, Single)#11Test mAP: 0.490
audio-classification-on-esc-50DyMN-L#5Top-1 Accuracy: 97.4PRE-TRAINING DATASET: AudioSet
audio-classification-on-fsd50kMN#2mAP: 65.6
audio-classification-on-fsd50kDyMN-L#4mAP: 65.5
audio-tagging-on-audiosetDyMN-L (Audio-Only, Single)#7mean average precision: 0.490
instrument-recognition-on-openmic-2018DyMN-L#1mean average precision: 0.855