CSTR-Edinburgh / merlin

This is now the official location of the Merlin project.
http://www.cstr.ed.ac.uk/projects/merlin/
Apache License 2.0
1.31k stars 440 forks source link

Network Type for Acoustic and Duration Models #477

Open kevinmchu opened 5 years ago

kevinmchu commented 5 years ago

I am running the run_full_voice.sh script in the egs/slt_arctic/s1 directory, but I am training on bdl instead of slt. In the demo paper, it says that the first recipe uses a feed-forward DNN architecture. However, when I synthesize new sentences, the following lines are output on the terminal when calculating acoustic and duration features.

Architecture:network_type has default value RNN Architecture:model_type has default value DNN

Does this mean that the model is actually deep recurrent rather than deep feed-forward?