Closed localaization closed 1 year ago
There is a lot of information that we could potentially get from: https://magenta.tensorflow.org/
I didn't know about it until now. I will take note about some interesting links: https://magenta.tensorflow.org/ddsp-vst-blog https://openreview.net/forum?id=B1x1ma4tDr¬eId=lcOwh022Vta https://github.com/magenta/mt3 https://archives.ismir.net/ismir2021/paper/000030.pdf https://openreview.net/pdf?id=iMSjopcOn0p
I've updated the documentation with the latest articles I've seen. I will meet Jaime tomorrow and leave this task unassigned for now.
Jaime and myself have been talking about this issue. We have red the A. Event-based Representation for Music from the document MuseMorphose, also another papers and pianogenie project. We think the pentagrom system could have one main advantage:
@isamu-isozaki this is the task, if you think it is solved feel free to mark it as done, otherwise assign it to me and back to in progress.
@localaization yup nice work! I think it's done. I think for machine learning, we just go through your format -> remi -> model which should do the job. Possibly midi in between
Description: We are evaluating the A. Event-based Representation for Music of the document MuseMorphose. We need to check with Jaime this method.
Documentation: From MuseMorphose and example midi file.zip
From This Time with Feeling: Learning Expressive Musical Performance The piano is used in this example
From SEQUENCE-TO-SEQUENCE PIANO TRANSCRIPTION WITH TRANSFORMERS
From MT3: MULTI-TASK MULTITRACK MUSIC TRANSCRIPTION
From Twitter
Piano mapping
Does not univocally map any written note.
Pentagrom mapping
It maps univocally any written note.
Solution
Q&A:
Definition of Done: To have an agreement whether or not section "A. Event-based Representation for Music" from MuseMorphose is a good approximation.