Dataset analysis

TTS (Text-to-Speech)
#1

Hi, I am struggling with training, still. So, I wondered- maybe there is something wrong with my dataset.
And I checked using AnalyzeDataset notebook

Selection_002
Selection_002967×746 80.9 KB

I wonder, if I did everything fine. Dataset is training, with poor result, however, it trains.
Have I done everything correct?

#2

You are populating items with the mailabs function (which is just one of several, plus you can if needs be write your own if your files/data are in a different structure) Using a suitable one is what the comment there is drawing attention to.

Are you sure that your files are in a format/directory structure that matches?

I’ve seen several cases where people set up their data in the structure for LJSpeech.

I’d suggest you Google both M-AI-Labs and LJSpeech to figure out if your structure matches them (or is some other structure) and make changes as appropriate, until the number of files is non-zero (ie that they’ve all been loaded correctly)

1 Like
#3

Yes, I did it in LJSpeech way, even names are the same, however only mailabs works without giving any error. I will check it again, and I will come back

#5

Now it’s working. The data path should be ‘more complicated’, as in the exemplary notebook.
Worth mentioning is that i had to rerun couple of times 7th block of commands. Before I had to do something with widgets: https://github.com/tqdm/tqdm/issues/394.

@erogol ,
according to the [notebook]> (https://github.com/mozilla/TTS/blob/master/dataset_analysis/AnalyzeDataset.ipynb):

/usr/local/lib/python3.6/dist-packages/ipykernel_launcher.py:18: TqdmDeprecationWarning: This function will be removed in tqdm==5.0.0
Please use tqdm.notebook.tqdm instead of tqdm.tqdm_notebook

[I couldn’t start a threat there]