New Language

[gaziway]

First of all thank you for releasing the codes.
I would like to know how difficult will be to do the training on a speakers data on a new language such as Turkish. As far as I sow during the generation step there is need for some kind of pronunciation dictionary. But what about pre-processing steps, Merlin and other tools, are they language agnostic. Thank you in advance

[adampolyak]

Yes, it is possible.

Preparing new data, requires 2 steps:

1. Extract phonemes - it possible to do so using https://github.com/bootphon/phonemizer. The documentation suggest that Turkish is supported.

2. Extract acoustic features - these features are agnostic to the language. You can extract using this script - just update the relevant paths to the tools directory downloaded in this repo.

[gaziway]

I am trying to process your answer, for which I thank you.
Let me summarize what I understand, so you could, please, correct me if needed.
Preparing new data should be done by extract_feats.py which as input accepts folders of txt and wav files.
Hence the next natural questions is how should one combine the steps you proposed in your answer with the extract_feats.py
One alternative is:

1. The content of the original text files should be replaced by their phonemes codes produced by phonemizer tool.
2. for extracting acoustic feature try to combine codes from second point with extract_feats.py

[adampolyak]

You can try to run extract_feats.py as usual and then simply update the generated npz:

save_dict = dict(numpy.load(npz_path))
save_dict['text_features'] = np.array(# run phonemizer here)
np.savez_compressed(new_npz_path, **save_dict)

[alex73]

I have own voice dataset like LJSpeech, with metadata.csv and wavs/*.wav files.
Also, I'm able to convert text into phonemes via phonemizer.
I started this script and it created: bap/*.bap, lf0/*.lf0, mgc/*.mgc files for each my wav file.

But what should be my next step for training ?