DavidBarbera / Few-Shot_Sound_Event_Detection_with_Prototypical_Neural_Networks

A study in using metric-based Meta-Learning for Sound Event Detection

Few-Shot Sound Event Detection with Prototypical Neural Networks

Study in using Metric-based Meta Learning for Sound Event Detection. It is based in the following two articles:

1. Wang, Y., Salamon, J., Bryan, N. J., & Pablo Bello, J. (2020). Few-Shot Sound Event Detection. ICASSP 2020
2. Snell, J., Swersky, K., & Zemel, R. (2017). Prototypical Networks for Few-shot Learning. Advances in Neural Information Processing Systems

Figure 1. Metric-based few-shot learning of a 5-way 2-shot model (left¹) and Prototypical Network (right²) of 3-way 5-shot model where a prototype for each class is created using its corresponding support set for each class.

Note: Works and Tested in my computer. However, as I am creating this repo open source there is still some work in progress which I will do in the following days. Requires:

1. Finish Readme
2. Installation instructions
3. Scripts to Download and preprocess SWC corpora to recreate experiment.

Highlight: Reproduces similar results as Wang et al. 2020 on SWC corpora for Prototypical Neural Networks despite using different split sets.