Maximizing Audio Event Detection Model Performance on Small Datasets Through Knowledge Transfer, Data Augmentation, And Pretraining: An Ablation Study

TL;DR

This paper investigates how knowledge transfer, data augmentation, and pretraining improve audio event detection on small datasets, demonstrating their individual contributions and proposing a smaller model that nearly achieves state-of-the-art results.

Contribution

It provides an ablation study analyzing the impact of different components on performance and introduces a compact model with competitive accuracy.

Findings

Knowledge transfer from ImageNet improves accuracy.

Pretraining on AudioSet enhances performance.

A smaller model achieves near SOTA results with fewer parameters.

Abstract

An Xception model reaches state-of-the-art (SOTA) accuracy on the ESC-50 dataset for audio event detection through knowledge transfer from ImageNet weights, pretraining on AudioSet, and an on-the-fly data augmentation pipeline. This paper presents an ablation study that analyzes which components contribute to the boost in performance and training time. A smaller Xception model is also presented which nears SOTA performance with almost a third of the parameters.