Silasimo/HubertFA-SynthGT

This checkpoint is a HubertFA model trained on our synthetic-singing dataset, SynthGT, for singing-oriented forced alignment experiments in our work "Synthesizing Ground Truth Phoneme Boundaries with SynthGT: A Synthetically Generated Solo-Singing Dataset". To learn more about this work, check out our GitHub repo, or the SynthGT repo here on Hugging Face. An associated paper is currently under review.

We release this checkpoint under a cc-by-nc-sa-4.0 license similar to the SynthGT dataset.

If you use this model, SynthGT, or other related resources, please cite:

bibtex
@unpublished{Antonisen2026SynthGT,
  author = {Silas Antonisen and Iván López-Espejo},
  title  = {Synthesizing Ground Truth Phoneme Boundaries with SynthGT: A Synthetically Generated Solo-Singing Corpus},
  note   = {Submitted to IEEE Transactions on Audio, Speech and Language Processing},
  year   = {2026}
}