Different max decoder steps in both tacotron and tacotron2

[JRMeyer]

>>> haqkiemdaim
[October 22, 2019, 8:07am]

I would like to know, inside layers/ , why they got different values
of max_decoder_steps in both tacotron.py and tacotron2.py ?

in layers/tacotron.py , the self.max_decoder_steps = 500

BUT

in layers/tacotron2.py , the self.max_decoder_steps = 1000

Which i assume MAYBE that could be the reason why i keep on getting the
message

warning: audio amplitude out of range, auto clipped.
| > Synthesizing test sentences
| > Decoder stopped with 'max_decoder_steps
| > Decoder stopped with 'max_decoder_steps
| > Decoder stopped with 'max_decoder_steps
| > Decoder stopped with 'max_decoder_steps

during training of Tacotron 2 but everything went fine on plain
Tacotron.

Any explanation for this max_decoder_steps values together with the
warnings ?

[This is an archived TTS discussion thread from discourse.mozilla.org/t/different-max-decoder-steps-in-both-tacotron-and-tacotron2]

[JRMeyer]

>>> erogol
[October 22, 2019, 9:36am]

you should really start to read some source code. It is the upper limit
for decoder to iterate. If stopnet does not predict right we don't like
to waste computation for going forever with the decoder.

[Archived Post]