search

choijeongsoo / lip2speech-unit

[Interspeech 2023] Intelligible Lip-to-Speech Synthesis with Speech Units
Other
25 stars 2 forks source link

Training subset #14

Closed longkhanh-fam closed 4 days ago

longkhanh-fam commented 2 weeks ago

Hi, Which subsets do you use to train in lrs2 and lrs3, only train-val subset or both train-val and pretrain? And what is the used ratio between subsets

choijeongsoo commented 4 days ago

Hi!

We used both trainval and pretrain subsets, excluding this validation set.

Regarding the ratio, approximately 1:10? For training LRS2 model, we used the full train and pretrain data from LRS2 dataset. For LRS3 model, we utilized unseen split of LRS3 dataset following SVTS.

longkhanh-fam commented 4 days ago

Yes, thank you for your response. So for training the LRS2, both pretrain and full training set of LRS2 are used. May I ask what exactly in the validation set of LRS2? Does it contain any pretrain samples like STVS?

choijeongsoo commented 4 days ago

We used official validation subset of LRS2 dataset, and it doesn't contain any pretrain samples if I remember correctly.

longkhanh-fam commented 4 days ago

Thanks @choijeongsoo, really appreciate your support ! I will close the issue here.