Training neural audio classifiers with few data

Jordi Pons, Joan Serrà, Xavier Serra
10/24/2018
cs.SDcs.AIcs.LG

Official code: jordipons/neural-classifiers-with-few-audio (60 stars, last commit 2019-02-01).

Abstract

We investigate supervised learning strategies that improve the training of neural network audio classifiers on small annotated collections. In particular, we study whether (i) a naive regularization of the solution space, (ii) prototypical networks, (iii) transfer learning, or (iv) their combination, can foster deep learning models to better leverage a small amount of training examples. To this end, we evaluate (i-iv) for the tasks of acoustic event recognition and acoustic scene classification, considering from 1 to 100 labeled examples per class. Results indicate that transfer learning is a powerful strategy in such scenarios, but prototypical networks show promising results when one does not count with external or validation data.

View on arXivView PDF

Code Implementations(1)

jordipons/neural-classifiers-with-few-audioOfficial100%

Training neural audio classifiers with few data − https://arxiv.org/abs/1810.10274

6012PythonNov 3, 20187 years ago

Similar Papers

Self-supervised Learning for Acoustic Few-Shot Classification

84% match2024

Rethinking CNN Models for Audio Classification

84% match1 repo1292020

Tuning In: Analysis of Audio Classifier Performance in Clinical Settings with Limited Data

83% match2024

Improving Audio Classification by Transitioning from Zero- to Few-Shot

83% match2025

Learning from Between-class Examples for Deep Sound Recognition

82% match1 repo952017

Self-Supervised Learning for Few-Shot Bird Sound Classification

82% match1 repo102023

Cite this paper

@article{pons2018training,
  title  = {Training neural audio classifiers with few data},
  author = {Jordi Pons and Joan Serrà and Xavier Serra},
  year   = {2018},
  eprint = {1810.10274},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1810.10274}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1810.10274.svg)](https://codeofpaper.com/paper/1810.10274)

Discussion