Closed zuowanbushiwo closed 2 years ago
Hi @seorim0 https://github.com/seorim0/DCCRN-with-various-loss-functions/blob/main/tools_for_loss.py#L206 Why use view here, should not be transpose? input x is (..., freq,time) and we need (..., n_mels, time) output, so (..., time, freq) dot (freq, n_mels) -> (..., n_mels, time)? thanks!
view
transpose
x
(..., freq,time)
(..., n_mels, time)
(..., time, freq)
(freq, n_mels) -> (..., n_mels, time)
Hi. What you understand is correct. There is no particular reason to use a view.
Hi @seorim0 https://github.com/seorim0/DCCRN-with-various-loss-functions/blob/main/tools_for_loss.py#L206
Why use
view
here, should not betranspose
? inputx
is(..., freq,time)
and we need(..., n_mels, time)
output, so(..., time, freq)
dot(freq, n_mels) -> (..., n_mels, time)
? thanks!