Making LJSpeech Tacotron2 model model say 'video'.

[bduvenhage]

Hi

This is possibly a question on TTS in general, but I'm fine tuning PyTorch Tacotron2 and I found that even the pre-trained LJSpeech model has trouble saying 'video'. I see the training and inference code supports arpabet, but I assume that the pre-trained models were trained on the plain text of LJSpeech and not phonetic sequences? I tried doing inference using arpabet sequences in curly brackets, but it doesn't work.

Do you have any recommendations for making the model say modern words that might not have been part of the LJSpeech corpus?

Thanks,