charsiu

/

zh_w2v2_tiny_fc_10ms

like 1

Cross-posting from the GitHub repo for visibility:
https://github.com/lingjzhu/charsiu/issues/44

Hi, and thanks for Charsiu — the forced aligner and the tone-marked
Mandarin phone inventory have been really useful.

A licensing question about the weights specifically. The GitHub repo
(lingjzhu/charsiu) carries an MIT LICENSE, but its text grants rights
over "the Software and associated documentation files", and the
pretrained weights are distributed separately here on the Hub. I checked
charsiu/zh_w2v2_tiny_fc_10ms, charsiu/en_w2v2_fc_10ms and
charsiu/zh_xlsr_fc_10ms — none of them has a LICENSE file, and none has
a model card with a license field, so the Hub shows no license for any
of them.

Could you confirm what license applies to the model weights themselves?
Are they intended to be covered by the repo's MIT license, or under
different terms?

I'm asking because I'd like to use the Mandarin checkpoint in a
commercial product. Given the paper states "Code and pretrained models
are available at [the GitHub repo]", my reading is that MIT was the
intent and the missing model-card metadata is simply an oversight — but
without an explicit statement, the conservative default assumption has
to be "all rights reserved", which unfortunately rules it out for
commercial use.

The concrete fix would be adding a license: field to the model cards'
YAML front matter (or a LICENSE file in these repos). That would settle
it permanently for everyone, not just me — I doubt I'm the first person
to have wondered.

Thanks for the work either way.