Published June 15, 2023 | Version v1
Other Open

ENHANCING CONFIDENCE AND SPECIFICITY IN SOUND EVENT DETECTION: EXPLORING THE IMPACT OF NON-TARGET TRAINING CLASSES pre_trained models, annotations and additional results.

  • 1. Politecnico di Milano; Universite de Lorraine, CNRS, Inria, Loria

Description

This repo contains the pre-trained models, annotations, and additional results related to the paper “ENHANCING CONFIDENCE AND SPECIFICITY IN SOUND EVENT DETECTION: EXPLORING THE IMPACT OF NON-TARGET TRAINING CLASSES.”

In all the files: 

  • BR stands for blind_rejection;
  • ALL stands for ALL-in;
  • AS stand for AudioSet;
  • proc stands for post-processed labels; 
  • soft stands for Softmax activation function, otherwise Sigmoid has been used. 

 

 

 

Files

additional_results.zip

Files (387.5 MB)

Name Size Download all
md5:98d8e88e4fb46306454a54da270b0c1a
527.6 kB Preview Download
md5:9283094b1b8ad9613d03235d8104d07f
2.3 MB Preview Download
md5:14eb79ab88014b955753630f461ed554
384.7 MB Preview Download