Implementation release timeline for Seq2seq piano transcription with transformers

80 views
Skip to first unread message

Richhiey Thomas

unread,
Oct 1, 2021, 12:56:12 PM10/1/21
to Magenta Discuss
Hi Magenta team,

I recently read this paper on seq2seq piano transcription with transformers. The results look great and I would like to recreate the experiments. The link to the implementation provided in the paper says that the code and pre-trained checkpoints will be made available soon. Could I please know by when this implementation will be released?

Thanks!

Best,
Richhiey

Sonnery Hugo

unread,
Oct 1, 2021, 1:21:42 PM10/1/21
to Magenta Discuss, richhi...@gmail.com
Incidentally, I have the same question and would be very interested in case you were planning on releasing the implementation !

Thanks !

Ian Simon

unread,
Oct 1, 2021, 2:11:38 PM10/1/21
to Sonnery Hugo, Magenta Discuss, richhi...@gmail.com
Hi all, we will be releasing our implementation of seq2seq piano transcription within a month or so.  The reason for the delay is that the model uses a JAX-based version of the T5 framework that is still in the process of being open-sourced.

-Ian

--
Magenta project: magenta.tensorflow.org
To post to this group, send email to magenta...@tensorflow.org
To unsubscribe from this group, send email to magenta-discu...@tensorflow.org
---
To unsubscribe from this group and stop receiving emails from it, send an email to magenta-discu...@tensorflow.org.

Curtis "Fjord" Hawthorne

unread,
Nov 17, 2021, 3:21:37 PM11/17/21
to Ian Simon, Sonnery Hugo, Magenta Discuss, richhi...@gmail.com
Just to update this thread in case you missed the announcement elsewhere, we've released the code and followup work on multi-instrument transcription in this blog post: https://g.co/magenta/mt3

-Fjord
Reply all
Reply to author
Forward
0 new messages