Incorrect vocab file generated in p2g mode.

[naok]

Hello

I tried recently added p2g mode as below.

g2p-seq2seq --train cmudict/cmudict.dict --model_dir p2g-model-nostress --cleanup --p2g

After training, I noticed strange vocabulary file was generated.
I suppose it should be identical with the one with g2p mode (PAD, EOS, phoneme symbols, characters), however it contains word spelling instead of characters.

Could you have a look if this vocab.g2p.zip sensible?

Also if you could provide a pre-trained p2g model, that would be helpful.

Thank you.

[b00f]

This is also happened to me. You can check the trained model here: https://github.com/tihu-nlp/tihudict