Finn interviews Composer and Machine Learning specialist Dr. Cheng-Zhi Anna Huang about the Music Transformer project at Google’s Magenta Labs. They discuss representations of music for machine learning, algorithmic music generation as a compositional aid, the JS Bach Google Doodle, how self-reference defines structure in music, and compare the musicality of different systems with example outputs.



Time Stamps




* [0:01:05] Introducing Dr. Anna Huang



* [0:03:43] JS Bach Google Doodle



* [0:12:52] Representations of musical information for machine learning



* [0:16:26] Music Transformer project

* [0:25:15] RNN algorithm music sample





* [0:25:45] ABA structure challenge for generative systems

* [0:30:30] Vanilla Transformer algorithm music sample 



* [0:32:07] Music Transformer algorithm music sample 





* [0:36:30] Self Reference Visualisation (see blog post)



* [0:43:27] Everyday music implications



* [0:48:10] What this work says about music

* [0:50:01] Music Transformer trained on Jazz Piano 






Show notes




* Recommended project:

* Blog post: Huang, C.Z.A., Simon, I., & Dinculescu, M. (2018, Dec 12). Music Transformer: Generating Music with Long-Term Structure [Blog Post]



* Paper: Huang, C.Z.A., Vaswani, A., Uszkoreit, J., Shazeer, N., Simon, I., Hawthorne, C., Dai, A.M., Hoffman, M.D., Dinculescu, M., & Eck, D. (2018) MUSIC TRANSFORMER: GENERATING MUSIC WITH LONG-TERM STRUCTURE on arXiv.org





* Interviewee:  Dr. Cheng-Zhi Huang at Google AI, on twitter @huangcza



* Google Doodle Celebrating JS Bach with AI harmonising melodies



* Related papers:

* Huang, C.Z.A., Cooijmans, T., Roberts, A., Courville, A., Eck, D. (2017). Coconet: Counterpoint by Convolution. ISMIR.



* Huang, C.Z.A., Cooijmans, T., Dinculescu, M., Roberts, A., & Hawthorne, C. (2019, Mar 20). Coconet: the ML model behind today’s Bach Doodle.



* Huang, C.Z.A., Hawthorne, C., Roberts, A., Dinculescu, M., Wexler, J., Hong, L., Howcroft, J. (2019). The Bach Doodle: Approachable music composition with machine learning at scale. ISMIR.






Credits



The So Strangely Podcast is produced by Finn Upham, 2018. Algorithmic music samples from the blog post Music Transformer: Generating Music with Long-Term Structure, and included under the principles of fair dealing. The closing music includes a sample of Diana Deutsch’s Speech-Song Illusion sound demo 1.