I guess it's a bad idea to forward inputs directly

Avoid relying on the order of the layers

Remove the metric because it wasn't too useful and it's buggy in
Keras 2.2.4.

Thanks to changeforan on Github

-train does data augmentation, cuts silence, and outputs the preemphasis
-test only outputs the features for resynthesis

David Rowe authored 6 years ago and Jean-Marc Valin committed 6 years ago

Signed-off-by: Jean-Marc Valin <jmvalin@jmvalin.ca>

David Rowe authored 6 years ago and Jean-Marc Valin committed 6 years ago

Signed-off-by: Jean-Marc Valin <jmvalin@jmvalin.ca>

David Rowe authored 6 years ago and Jean-Marc Valin committed 6 years ago

Signed-off-by: Jean-Marc Valin <jmvalin@jmvalin.ca>

Without the 0.1 bias, the rounding error could cause an offset of -1