One model trained on all data
#1
by
pplantinga
- opened
There's ten models here and its not clear which one is best, could you train one model on all the data?
These models come with the paper: https://arxiv.org/abs/2406.16128 . Each model is trained on 9 out of 10 folds and tested on the missing one (folds available here: https://github.com/K-STMLab/SSL4PR/tree/main/pcgita_splits)
If you want model available in Table 3 instead (trained on the whole s-PC-GITA and tested on e-PC-GITA), those are not available in this space, I can upload on the hub if needed.
I just uploaded the two requested models. You can find the models trained on the full s-PC-GITA dataset below
- WavLM-based https://huggingface.co./morenolq/SSL4PR-wavlm-base-full
- HuBERT-based https://huggingface.co./morenolq/SSL4PR-hubert-base-full
Hope this helps!
Thanks! Will check out :)
pplantinga
changed discussion status to
closed