TVCGMM Pretrained Model Checkpoints

To run inference in our interactive demo and experiments it suffices to download our pre-trained model checkpoints. The downloaded .zip file contains a 40000.pth.tar and a used_config.yaml which should both be placed together in the directory output/ckpt/.

We provide checkpoints for the three supported datasets LJSpeech, VCTK and LibriTTS and our three models FastSpeech 2 (baseline), TVC-GMM [k=1] and TVC-GMM [k=5].