How long does it take for 200 epochs?

Somewhere around 1 to 1.5hrs

also depends on the dataset and the GPU

Ohhh okkk

This is my first time using the newer versions of tacotron 2 and I’m still kind of used to the older versions because I made a model back in like January and it turned out to be terrible so I’m trying to learn my mistakes and stuff

Btw this is epoch 51

That doesn't look good

You've gotta work on your dataset

can someone answer my eriir

error

Make sure the audio clips are clean and also make sure the clips are no longer than 11 secs

Here I am at epoch 50

RuntimeError Traceback (most recent call last) in () 56 57 device = "cuda:0" ---> 58 taco.from_pretrained(warm_start_path="TTSModel.pt", device="cpu") 59 if device != "cpu": 60 taco.cuda() 1 frames /usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py in load_state_dict(self, state_dict, strict) 1496 if len(error_msgs) > 0: 1497 raise RuntimeError('Error(s) in loading state_dict for {}:\n\t{}'.format( -> 1498 self.class.name, "\n\t".join(error_msgs))) 1499 return _IncompatibleKeys(missing_keys, unexpected_keys) 1500 RuntimeError: Error(s) in loading state_dict for Tacotron2: Unexpected key(s) in state_dict: "speaker_embedding.weight", "gst_lin.weight", "gst_lin.bias".

Got this in HIFIGAN Pipeline

dont use a finalized model

Has anyone been trying to make end-to-end Gan speech synthesizer that does not have to upsample but uses high quality raw audio?

google did that years ago with wavenet but switched away from it because it was hella slow there's also vits which is just glow-tts + hifi-gan trained end-to-end, i tried it but it kept overfitting and mispronouncing stuff even with hours of data

Alright, so this is currently slow but in upcoming years will become more and more popular

Do you have any article accordaing to that google thing with wavenet?

that was years ago

before even tacotron i think

i think they use it as a vocoder nowadays

,,or not? hm

how many checkpoints does hifi-gan is good enough?

how would i know if it good?

Epoch 58, still unsure when would be a good time to create my model from this training

that is an extremely deranged graph

you did disable arpabet right?