See you guy have a great work on AlignTTS. I have some concerns about training. I have tried to reimplement MDN loss, Viterbi algorithm Tensorflow version following your Torch code, and training phase 0. The loss is similar to yours. I wanna check that log_prob and alpha actually look like? Could you guys give me some pictures of log_prob and alpha? Here is my figure about loss, log_prob, and alpha
See you guy have a great work on AlignTTS. I have some concerns about training. I have tried to reimplement MDN loss, Viterbi algorithm Tensorflow version following your Torch code, and training phase 0. The loss is similar to yours. I wanna check that log_prob and alpha actually look like? Could you guys give me some pictures of log_prob and alpha? Here is my figure about loss, log_prob, and alpha