Producing Symbolic Music with High-quality-Grained Inventive Management


Obtain a PDF of the paper titled FIGARO: Producing Symbolic Music with High-quality-Grained Inventive Management, by Dimitri von R”utte and three different authors

Obtain PDF
HTML (experimental)

Summary:Producing music with deep neural networks has been an space of lively analysis in recent times. Whereas the standard of generated samples has been steadily rising, most strategies are solely in a position to exert minimal management over the generated sequence, if any. We suggest the self-supervised description-to-sequence process, which permits for fine-grained controllable technology on a world degree. We achieve this by extracting high-level options in regards to the goal sequence and studying the conditional distribution of sequences given the corresponding high-level description in a sequence-to-sequence modelling setup. We prepare FIGARO (FIne-grained music Technology by way of Consideration-based, RObust management) by making use of description-to-sequence modelling to symbolic music. By combining realized excessive degree options with area data, which acts as a robust inductive bias, the mannequin achieves state-of-the-art leads to controllable symbolic music technology and generalizes properly past the coaching distribution.

Submission historical past

From: Dimitri von Rütte [view email]
Wed, 26 Jan 2022 13:51:19 UTC (590 KB)
Tue, 1 Feb 2022 12:33:01 UTC (590 KB)
Tue, 1 Mar 2022 09:36:11 UTC (590 KB)
Thu, 22 Feb 2024 10:34:18 UTC (1,133 KB)

Supply hyperlink


Please enter your comment!
Please enter your name here