%0 Conference Paper %1 pmlr-v119-choi20b %A Choi, Kristy %A Hawthorne, Curtis %A Simon, Ian %A Dinculescu, Monica %A Engel, Jesse %B Proceedings of the 37th International Conference on Machine Learning %D 2020 %E III, Hal Daumé %E Singh, Aarti %I PMLR %K controllable generation %P 1899--1908 %T Encoding Musical Style with Transformer Autoencoders %U https://proceedings.mlr.press/v119/choi20b.html %V 119 %X We consider the problem of learning high-level controls over the global structure of generated sequences, particularly in the context of symbolic music generation with complex language models. In this work, we present the Transformer autoencoder, which aggregates encodings of the input data across time to obtain a global representation of style from a given performance. We show it is possible to combine this global representation with other temporally distributed embeddings, enabling improved control over the separate aspects of performance style and melody. Empirically, we demonstrate the effectiveness of our method on various music generation tasks on the MAESTRO dataset and a YouTube dataset with 10,000+ hours of piano performances, where we achieve improvements in terms of log-likelihood and mean listening scores as compared to baselines.

@inproceedings{pmlr-v119-choi20b, abstract = {We consider the problem of learning high-level controls over the global structure of generated sequences, particularly in the context of symbolic music generation with complex language models. In this work, we present the Transformer autoencoder, which aggregates encodings of the input data across time to obtain a global representation of style from a given performance. We show it is possible to combine this global representation with other temporally distributed embeddings, enabling improved control over the separate aspects of performance style and melody. Empirically, we demonstrate the effectiveness of our method on various music generation tasks on the MAESTRO dataset and a YouTube dataset with 10,000+ hours of piano performances, where we achieve improvements in terms of log-likelihood and mean listening scores as compared to baselines.}, added-at = {2023-05-19T11:42:29.000+0200}, author = {Choi, Kristy and Hawthorne, Curtis and Simon, Ian and Dinculescu, Monica and Engel, Jesse}, biburl = {https://www.bibsonomy.org/bibtex/26e790091daacd6c429dc1ced86aa6ac9/alex_h}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, description = {Encoding Musical Style with Transformer Autoencoders}, editor = {III, Hal Daumé and Singh, Aarti}, interhash = {7c9903f0370a58a6f78996c383495e2b}, intrahash = {6e790091daacd6c429dc1ced86aa6ac9}, keywords = {controllable generation}, month = {13--18 Jul}, pages = {1899--1908}, pdf = {http://proceedings.mlr.press/v119/choi20b/choi20b.pdf}, publisher = {PMLR}, series = {Proceedings of Machine Learning Research}, timestamp = {2023-05-19T11:42:29.000+0200}, title = {Encoding Musical Style with Transformer Autoencoders}, url = {https://proceedings.mlr.press/v119/choi20b.html}, volume = 119, year = 2020 }