the notebook uses 11.2 so i think 11.3 is good

sim

ok

ok its done what do i run for train.py?

what hparams do i need

see hparams.py

ok

do i edit hparams.py?

so that it saves

no

read it and make a command line argument for train.py

so --hparams=training_files=filelists/files.txt

or with quotes

--hparams=training_files="filelists/files.txt"

without quotes

ok

hparams are separated by coma so you can multiple arguments

python train.py --output_directory=outdir --log_directory=logdir --hparams=training_files=filelists/files.txt,validation_files=filelists/files.txt

do i need any more params

is batch size too big default

you need to also specify warm start and check point path to pre trained model

also experiment with batch size until pytorch stops complaining about memory

if it doesn't first try good

what is warm start and checkpoint path

check point path is where it is warm starting from

hmm

warm start can ignore some layers iirc

so what value do i put for checkpoint_path

download the tacotron 2 pretrained from nvidia on the readme

and then put it in the tacotron 2 directory or something