Currently, the symbol for compound boundaries in the Braxen lexicon is not retained when converted to the WS format. However, it (#) was used when training the TTS models, and the G2P DeepPhonemizer model produces them.
It could be nice to use the exact same symbol set for inference as for training. (The boundary symbol is silent, so it cannot matter that much. No big deal.)
Currently, the symbol for compound boundaries in the Braxen lexicon is not retained when converted to the WS format. However, it (#) was used when training the TTS models, and the G2P DeepPhonemizer model produces them.
It could be nice to use the exact same symbol set for inference as for training. (The boundary symbol is silent, so it cannot matter that much. No big deal.)