ISMIR 2018 - Paper Overviews

This year’s ISMIR was great as ever, this time featuring

• lots of deep learning - I suspect since it became much more easy to use with recently developed libraries
• lots of new, and surprisingly large, datasets (suited for the new deep learning era)
• and a fantastic boat tour through Paris!

Here are some papers that caught my attention:

Deep Learning Papers

Deep Learning for Audio-Based Music Classification

This paper by Choi et al. presents a comprehensive comparison of different deep learning architectures for music classification tasks. They compare CNNs, RNNs, and hybrid architectures on various datasets and show that CNNs generally perform best for this task.

Music Source Separation with Deep Neural Networks

A nice overview paper by Stöter et al. on recent advances in music source separation using deep learning. They provide a good comparison of different approaches and discuss the challenges in this field.

Traditional MIR Papers

Beat Tracking with Deep Learning

This paper by Böck et al. shows how to use deep learning for beat tracking, achieving state-of-the-art results on standard datasets.

Dataset Papers

MUSDB18: A Dataset for Music Source Separation

A new large-scale dataset for music source separation by Rafii et al. This dataset will be very useful for future research in this area.

Conclusion

ISMIR 2018 showed that deep learning has become the dominant approach in music information retrieval. The quality of papers was generally very high, and there were many interesting discussions about the future of the field.