search

NVIDIA / OpenSeq2Seq

Toolkit for efficient experimentation with Speech Recognition, Text2Speech and NLP
https://nvidia.github.io/OpenSeq2Seq
Apache License 2.0
1.54k stars 371 forks source link

tacotron gst: style embdeddings #462

Open gerasimoss opened 5 years ago

gerasimoss commented 5 years ago

I am running Tacotron GST with MAILABS and I would like to get the style embeddings and the weights of the style tokens during inference. How could I do it?

blisc commented 5 years ago

Are you interested in gst_embedding and token_embeddings from _embed_style()?

You would have to provide those tensors to sess.run(); eg sess.run([gst_embedding, token_embeddings). The easiest way would probably to change get_interactive_infer_results() such that the fetches variable contains the relevant tensors of interest.

You can see #296 and #361 for tips on how to enable interactive infer for tacotron_gst