Automatic Tagging Using Deep Convolutional Neural Networks

By Keunwoo Choi, György Fazekas, Mark B. Sandler. International Society for Music Information Retrieval Conference, 2016

351 citations recorded by the source corpus.

We present a content-based automatic music tagging algorithm using fully convolutional neural networks (FCNs). We evaluate different architectures consisting of 2D convolutional layers and subsampling layers only. In the experiments, we measure the AUC-ROC scores of the architectures with different complexities and input types using the MagnaTagATune dataset, where a 4-layer architecture shows state-of-the-art performance with mel-spectrogram input. Furthermore, we evaluated the performances of the architectures with varying the number of layers on a larger dataset (Million Song Dataset), and found that deeper models outperformed the 4-layer architecture. The experiments show that mel-spectrogram is an effective time-frequency representation for automatic tagging and that more complex models benefit from more training data.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Signal Processing

View Paper: https://www.semanticscholar.org/paper/b9ba8c4a00f5ee43e768db2acc8b56f017176f3e

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

Deep Learning for Audio Signal Processing (2019, methodology): “[110] use a FCN of 3×3 convolutions interleaved with max-pooling such that a 29-second log-mel spectrogram is reduced to a 1×1 feature map and classified.”
The MTG-Jamendo Dataset for Automatic Music Tagging (2019, methodology): “As a simple baseline experiment, we tested a well-known music auto-tagging model proposed by Choi et al. (2016).”
Wavelet Networks: Scale Equivariant Learning From Raw Waveforms (2020, methodology): “MAXPOOL: 5X1 FLATTEN 196× 6→ 1152 FC: [1152, 96] FC: [96, 48] FC: [48, 10] SOFTMAX Table 3: Proposed wavelet network variant of the 50999-1DCNN of Abdoli et al.”
Multi-level attention model for weakly supervised audio classification (2018, methodology): “Recently, neural network based methods including fully connected neural networks [16], convolutional neural networks (CNN) [17] a r X i v : 1803 .”
Convolutional recurrent neural networks for music classification (2016, methodology): “The widths of layers are based on [1] for Conv1D and [2] for Conv2D.”
Audio Set Classification with Attention Model: A Probabilistic Perspective (2017, methodology): “Maximum selection corresponds to a global max pooling layer [12] in a convolutional neural network.”
Multi-Level and Multi-Scale Feature Aggregation Using Pretrained Convolutional Neural Networks for Music Auto-Tagging (2017, methodology): “We filtered out the tags and used most frequently labeled 50 tags for both MTAT and MSD, following the previous work [2], [4].”
Sample-level Deep Convolutional Neural Networks for Music Auto-tagging Using Raw Waveforms (2017, methodology): “We ﬁltered out the tags and used most frequently labeled 50 tags in both datasets, following the previous work [11], [14,15] 1 .”
PANNs: Large-Scale Pretrained Audio Neural Networks for Audio Pattern Recognition (2019, methodology): “Log mel spectrograms are usually used as input to CNNs [32], [19].”
Evaluation of CNN-based Automatic Music Tagging Models (2020, methodology): “[1].”
Training general-purpose audio tagging networks with noisy labels and iterative self-verification (2018, background): “Naturally, a large number of similar architectures and neural networks have proven to be powerful tools in this setting [6, 7, 8, 9, 10, 11].”