Trainingless Adaptation of Pretrained Models for Environmental Sound Classification

TL;DR

This paper introduces a trainingless domain adaptation method for pretrained environmental sound classification models, improving accuracy without additional training or heavy computational resources.

Contribution

It proposes a novel trainingless adaptation approach using time-frequency structure recovery and frequency filtering, reducing reliance on resource-intensive fine-tuning.

Findings

Achieved a 20.40% accuracy improvement on ESC-50 dataset.

Demonstrated effectiveness of trainingless adaptation over conventional methods.

Abstract

Deep neural network (DNN)-based models for environmental sound classification are not robust against a domain to which training data do not belong, that is, out-of-distribution or unseen data. To utilize pretrained models for the unseen domain, adaptation methods, such as finetuning and transfer learning, are used with rich computing resources, e.g., the graphical processing unit (GPU). However, it is becoming more difficult to keep up with research trends for those who have poor computing resources because state-of-the-art models are becoming computationally resource-intensive. In this paper, we propose a trainingless adaptation method for pretrained models for environmental sound classification. To introduce the trainingless adaptation method, we first propose an operation of recovering time--frequency-ish (TF-ish) structures in intermediate layers of DNN models. We then propose the trainingless frequency filtering method for domain adaptation, which is not a gradient-based optimization widely used. The experiments conducted using the ESC-50 dataset show that the proposed adaptation method improves the classification accuracy by 20.40 percentage points compared with the conventional method.