all my files this run are equalized and i made extra sure everything else was ok with the reference audio and transcript

Of course That's just the quality of the preview audio TensorBoard uses Griffin Lim instead of HiFi-GAN as the neural vocoder for generating preview outputs

oh ok. i wasn't sure. ty

Can you explain this error? I cant screenshot for some reason but here (the notebook is my grimsalad): /content/tacotron2/text/__init__.py:74: SyntaxWarning: "is not" with a literal. Did you mean "!="? return s in _symbol_to_id and s is not '_' and s is not '~' /content/tacotron2/text/__init__.py:74: SyntaxWarning: "is not" with a literal. Did you mean "!="? return s in _symbol_to_id and s is not '_' and s is not '~' --------------------------------------------------------------------------- AttributeError Traceback (most recent call last) in 396 397 # ---- DEFAULT PARAMETERS DEFINED HERE ---- --> 398 hparams = create_hparams() 399 model_filename = 'current_model' 400 hparams.training_files = "filelists/clipper_train_filelist.txt" /content/tacotron2/hparams.py in create_hparams(hparams_string, verbose) 6 """Create model hyperparameters. Parse nondefault from given string.""" 7 ----> 8 hparams = tf.contrib.training.HParams( 9 ################################ 10 # Experiment Parameters # AttributeError: module 'tensorflow' has no attribute 'contrib'

how long need to train a pipeline tacotron?

hello

any help guys

Always remember to use the notebooks on #841437191073955920

They're constantly updated

I'm getting this error in the synthesis notebook, what's going on?

A question. Would training a dataset with 30 wavs in pipeline be recommended? I could've sword someone said that it's mostly used for that.

Yo, sorry for the ping but you can train a model in Pipeline with 4 wavs that are approximately 5 seconds long. You just need to set your batch size to 1, and learning rate to 0.0002

Oh okay. Should that work with around 30 wavs too?

OK with 30 wavs, I have no idea. You might need to ask hecko since he basically made the notebook.

Alrighty.

@hecko Do you know what settings for training a model with 30 wavs in your pipeline notebook would you recommend? (Sorry to ping btw)

I'm also getting a list index out of range error after uploading the dataset on drive. What am I doing wrong?

anyone getting this

batch size like 1 or 2 maybe

also do u guys think waveglow produces a more natural sound than hifi gan i know hifi is clearer

no transcript found

I did put the list.txt in the zip file

Or do I need to rename it to transcript?

It's my first time using your pipeline notebook btw lol

show me what it looks like

All I did was upload a zip file containing list.txt and wavs folder from 0 to 29. Btw, I'm doing this for someone.

I did exactly what the instruction said and it didn't work out.

It looks like this btw

Mickey Mouse Needs A Retrain