[FastPitch 1.1] training from phonemic data

[Nistrian]

Good afternoon. I noticed that the FastPitch model can train from phonemes. However, not all words are phonemicized when using this feature, as some of them are missing from the cmudict. At the same time, I found g2p_en https://pypi.org/project/g2p-en/, which successfully copes with its task. In this regard, the question arose: why do not you use it? Are there any pitfalls associated with learning from fully phonemic data?