Amit Moryossef

cool. the reason `zst` saves space, I think, is that we have many sequences of 0 and many sequences of 1. a `float16` storage option would allow to shrink by...

Awesome, thanks! I think an automatic update to `readthedocs` is still missing, and that would be it for this issue.

Hi! there is no jitter normalization function. If you want to work with poses, and get more consistent results, consider post-processing using something like https://omnimotion.github.io/ - I came up with...

1. I am not sure i understand the change. 2. if confidence is 0, a mask is created https://github.com/sign-language-processing/pose/pull/161/files#diff-b2fa5baab13397a32417b3f212c9429c3398ddf14222e50a8eafa2d1af5f379fR50

Initial thoughts > A SMPL-Body contains a 3D mesh, a skeleton rig, pose blendshapes and dynamic blendshapes. A SMPL-Body can be shared in the form of OBJ file(s), animated FBX...

Figma link: https://www.figma.com/design/wB26N3qIODgxHYq9zhRUUv/sign.mt-about?m=auto&t=dFTiYvIWI4ZMy3dI-6

- did it prompt you for audio permissions, and you allowed them? - it might be that in the manifest we don't correctly declare audio permissions

Thanks for investigating, I'm not sure I can assist on this. There should be support for the `SpeechRecognition` web API. 

You are correct, our build target is: https://github.com/sign-language-processing/pose/blob/master/src/js/pose_format/tsconfig.json#L10-L11 If you would like to create multiple targets, that is something that I would happily merge.

Thanks @kuaashish . point still stands - i was using 0.10.21 for the `tasks-text` package (works from npm) and now it is trying to upgrade to an older version `20230920.0.0`...