keonlee9420 / comprehensive-tacotron2 Goto Github PK

Hi @keonlee9420

How to install hifigan package in model.py file

Best regards,
PeterPham

@keonlee9420 Thanks for your great work of this repo!

However, I have some questions about this model:

1. I have used the pretrained model to get good performance on single speaker (LJSpeech), but can't synthesize speech on multi-speaker, becasue I dont have the speaker embedding on VCTK. Could you upload some speaker embedding .npy files such as the default speaker "p225" ? I am downloading VCTK but expect it to take a day.
2. Have you tested the performance on the LibriTTS dataset?

Hi @keonlee9420
Do you know how to solve this issue?

raceback (most recent call last):
File "D:\C\Comprehensive-Tacotron2-main\preprocess.py", line 28, in
main(preprocess_config)
File "D:\C\Comprehensive-Tacotron2-main\preprocess.py", line 13, in main
preprocessor.build_from_path()
File "D:\C\Comprehensive-Tacotron2-main\preprocessor\vctk.py", line 168, in build_from_path
self.divide_speaker_by_gender(self.in_dir), filename="spker_embed_tsne.png"
File "D:\C\Comprehensive-Tacotron2-main\utils\tools.py", line 314, in plot_embedding
data_y = np.array([gender_dict[spk_id] == 'M' for spk_id in embedding_speaker_id], dtype=np.int)
File "D:\C\Comprehensive-Tacotron2-main\utils\tools.py", line 314, in
data_y = np.array([gender_dict[spk_id] == 'M' for spk_id in embedding_speaker_id], dtype=np.int)
KeyError: 'preprocessed_data\VCTK\spker_embed\p225'

The link https://github.com/keonlee9420/Comprehensive-Tacotron2/blob/main indicated in the Readme shows a page not found 404.

i am from china, and find your program is wonderful.
But i don't know how to train a model for chinese.
Can you give some guidance?
very greatful.

Called in the code "speakers.json" ,but I didn't find this file,I hope you can answer this question, thank you

RuntimeError: Input tensor at index 1 has invalid shape [13, 432, 154], but expected [13, 432, 183]

I try training my own model but batch size 64 does not fit into the GPU memory. I wonder what hardware did you use for your own training.