Randomly weighted CNNs for (music) audio classification
This work addresses the problem of efficient feature extraction for music audio classification, offering an incremental approach by adapting computer vision insights to audio tasks.
The authors investigated whether randomly weighted convolutional neural networks (CNNs) can serve as effective feature extractors for music audio classification, finding that the choice of architecture significantly impacts classification accuracy, with specific architectures achieving competitive results without training.
The computer vision literature shows that randomly weighted neural networks perform reasonably as feature extractors. Following this idea, we study how non-trained (randomly weighted) convolutional neural networks perform as feature extractors for (music) audio classification tasks. We use features extracted from the embeddings of deep architectures as input to a classifier - with the goal to compare classification accuracies when using different randomly weighted architectures. By following this methodology, we run a comprehensive evaluation of the current deep architectures for audio classification, and provide evidence that the architectures alone are an important piece for resolving (music) audio problems using deep neural networks.