Is it okay to not have Libri data in training_list.txt when training Libri data?

[Moon-sung-woo]

I'm trying to train a model for multi-speaker.
When I look at train_data from config_libits.yml, this is using the same training_list.txt used for Ljspeech learning.
There are Libritts data in OOD_texts.txt, but it doesn't seem to affect learning with multi-speaker because only text is extracted.
Do I need to create a new filelist to learn Libri dataset? And edit config_libritts.yml?
Or should I write the given config_libbits.yml?
Thank you.

[lucasjinreal]

what do u mean

[Moon-sung-woo]

@lucasjinreal
In my opinion, to train Libri data, there should be Libri data in training_list,text, but there is none.
OOD data only has text for libri data.

[lucasjinreal]

OOD doesn't need wav data

[Moon-sung-woo]

@lucasjinreal
I know.
So I think Libri dataset should be in training_list.txt in config_libits.yml.
Currently training_list.txt has only data for LJspeech.

[martinambrus]

I guess this would be a bit of a de-zombification of this but rather late than never :)
the reason why train_list.txt does not contain data from Libri is because it's set up to train LJSpeech dataset
to train LibriTTS, you will need to replace train_list.txt and val_list.txt with data from LibriTTS
also, as previously mentioned, OOD texts don't need to have corresponding audio files - these are texts from outside of the training dataset (so like news articles, research studies, audiobooks etc.) and just need to be updated if you change the trained language (or want to train / validate the model on a specific dictionary)