Thank you for releasing TCSinger 2. The zero-shot style-transfer results are impressive, particularly the bel canto examples on the demo page.
I'd like to evaluate the model on Western operatic repertoire for an academic research project. The repo provides the training code and the HIFI-GAN and FLAN-T5 dependencies, but I didn't find a released inference checkpoint. Is a pretrained TCSinger 2 checkpoint available for research use, so I can run inference directly rather than training a base model from scratch?
If not, is training on GTSinger the intended path to a working inference setup, and are there any notes on the compute required? Happy to cite TCSinger 2 in any resulting work. Thank you.
Thank you for releasing TCSinger 2. The zero-shot style-transfer results are impressive, particularly the bel canto examples on the demo page.
I'd like to evaluate the model on Western operatic repertoire for an academic research project. The repo provides the training code and the HIFI-GAN and FLAN-T5 dependencies, but I didn't find a released inference checkpoint. Is a pretrained TCSinger 2 checkpoint available for research use, so I can run inference directly rather than training a base model from scratch?
If not, is training on GTSinger the intended path to a working inference setup, and are there any notes on the compute required? Happy to cite TCSinger 2 in any resulting work. Thank you.