rhasspy
Add support for Raw Phoneme Input and Other Phonemes
These changes update Piper to add a new phoneme type "phonemes" that a model can be configured with signifying it accepts raw phonemes as input and a phonemizer should not be used.
The phonemes can now also be defined as multi-character (1-3) ASCII which adds support for other phonetic alphabets such as Sampa and Arpabet,
The changes from #403 have been added to allow the --phoneme-input flag to be used in addition to the model config.
Some of the code from the piper_phonemize library relating to converting phonemes to IDs has been moved back into the piper application.
The rational for these changes is it means Piper can be used with a custom phonemizer working to preprocess sentences and without the dependency on the espeak library. On the python side, the espeak library is not needed to be even installed, on the C++ side it is still linked as it's a little complicated to disentangle it.
This relates to the piper phonemizer issue: https://github.com/rhasspy/piper-phonemize/issues/17
