gate outputs?
JRMeyer opened this issue · 5 comments
there are many mentions of gate
in the code, for example:
What is this gate
, and what does it do?
also, here are gate_targets
:
gate output of decoder is a flag indicating whether the decoding process has reach the last step.
If the decoding has finished, gate output should be >0.5, otherwise, it should be < 0.5.
The second paragraph of Section-III-D of our paper says:
In order to end the acoustic feature sequences generated
by the seq2seq decoder at the conversion stage, the hidden
state of the seq2seq decoder at each frame is projected to
a scalar followed by sigmoid activation to predict whether
current frame is the last frame in an utterance. Accordingly,
a cross entropy loss $L_{ED}$ is defined for this prediction at the
training stage.
I think that gate
should be changed to another word, because gate_output
is ambiguous between "the outputs of a gate" and "gate the outputs", and the word gate
is overloaded in machine learning.
how about is_final_frame
?
Yes, it's a good suggestion and I think it'll be more clear.
By the way, the name "gate_output" actually inherited from the Tacotron 2 repo of NVIDIA.