An empirical study of weakly supervised audio tagging embeddings for general audio representations

Abstract

We study the usability of pre-trained weakly supervised audio tagging (AT) models as feature extractors for general audio representations. We mainly analyze the feasibility of transferring those embeddings to other tasks within the speech and sound domains. Specifically, we benchmark weakly supervised pre-trained models (MobileNetV2 and EfficientNet-B0) against modern self-supervised learning methods (BYOL-A) as feature extractors. Fourteen downstream tasks are used for evaluation ranging from music instrument classification to language classification. Our results indicate that AT pre-trained models are an excellent transfer learning choice for music, event, and emotion recognition tasks. Further, finetuning AT models can also benefit speech-related tasks such as keyword spotting and intent classification.

arXiv commentOdyssey 2022

References in corpus (9)

• mixup: Beyond Empirical Risk Minimization
• SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition
• VoxCeleb: a large-scale speaker identification dataset
• Speech Commands: A Dataset for Limited-Vocabulary Speech Recognition
• Neural Audio Synthesis of Musical Notes with WaveNet Autoencoders
• Towards duration robust weakly supervised sound event detection
• HEAR: Holistic Evaluation of Audio Representations
• Multi-Format Contrastive Learning of Audio Representations
• Multimodal Self-Supervised Learning of General Audio Representations