search

CODEJIN / NaturalSpeech2

MIT License
140 stars 15 forks source link

Metadata.csv questions #9

Open CHK-0000 opened 8 months ago

CHK-0000 commented 8 months ago

(s) c:\NaturalSpeech2>python C:\NaturalSpeech2\Pattern_Generator.py -lj C:\NaturalSpeech2\dataset\train -hp C:\NaturalSpeech2\Hyper_Parameters.yaml C:\Users\user.conda\envs\s\lib\site-packages\torch\nn\utils\weight_norm.py:28: UserWarning: torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm. warnings.warn("torch.nn.utils.weight_norm is deprecated in favor of torch.nn.utils.parametrizations.weight_norm.") Traceback (most recent call last): File "C:\NaturalSpeech2\Pattern_Generator.py", line 1443, in lj_paths, lj_text_dict, lj_pronunciation_dict, lj_speaker_dict, lj_emotion_dict, lj_language_dict, lj_gender_dict = LJ_Info_Load(path= args.lj_path) File "C:\NaturalSpeech2\Pattern_Generator.py", line 1042, in LJ_Info_Load for line in open(os.path.join(path, 'metadata.csv').replace('\', '/'), 'r', encoding= 'utf-8-sig').readlines(): FileNotFoundError: [Errno 2] No such file or directory: 'C:/NaturalSpeech2/dataset/train/metadata.csv'

I'm getting the above error, where can I get the metadata.csv? Please let me know the format of the data. Thanks.

CODEJIN commented 8 months ago

Dear @ksl103177 ,

Hello. Is the 'C:\NaturalSpeech2\dataset' you specified where the LJSpeech dataset is located? metadata.csv is located at the root of the dataset. To be exact, the unzipped LJSpeech dataset consists of one folder and two files as follows.

-wavs -metadata.csv -README

If there is no metadata.csv in your LJSpeech dataset, download it again from here and unzip it.

Best regards.

CHK-0000 commented 8 months ago

Oh, thank you, I have another question. In the Hyper_Parameters.yaml file, should I use the Path: 'C:\NaturalSpeech2\LJSpeech-1.1\wavs' path in Train_pattern as the train data and the data in Eval_Pattern should be in the same format?

CODEJIN commented 2 months ago

No. The train and eval paths are the locations where the training data is generated. Pattern_Generator.py outputs a completely separate pickled file from the dataset and inputs it into the specified folder. Therefore, the train and eval paths should not be LJSpeech-1.1 but rather a new folder should be designated.