How Jasper and QuartzNet process residual connections ? #2440
Unanswered
Arsham-boredom
asked this question in
Q&A
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
To proceed Residual connection, each SubBlock output tensor add to residual tensor, which has 1x1Conv followed by BatchNorm.
these tensors need to be singleton, any mismatch in their shape will raise "mismatch" error.
in resnet, inputs get resized to some fixed size (32x32 for example) and kernel and strides are computed manually so block outputs and residual have same tensor shape.
here is my question, ASR models' inputs may vary in sequence length (5s up to 25s), they are NOT fixed.
how these models handle residual connection in this case?
Beta Was this translation helpful? Give feedback.
All reactions