[Bug] Phoneme extraction with punctuations is wrongly delimited

Describe the bug Punctuations in extracted phonemes are delimited wrongly.

For instance the sentence tuː foːɹ paʊndz , bʌt hɛviɚ aɪɚnz , should be tuː foːɹ paʊndz, bʌt hɛviɚ aɪɚnz,

So punctuations do not need a space preceding them.

I think the current implementation causes unnatural silences in the trained models.

[synesthesiam]

I’m working on it! I have this pesky job that keeps taking my time 😉

This is taking longer than expected since I’m adding it and preliminary SSML support at this same time. It didn’t seem worth it to me to redo the existing gruut tokenizer (to add proper whitespace preservation) only to scrap it later for SSML.

I will be completing the changes to gruut this week, and my goal is to have it integrated and tested with 🐸 TTS by 1 Oct 👍

[synesthesiam]

I’m reworking parts of gruut’s tokenization pipeline to preserve whitespace. I’ll delay updating the current pull request until these changes are in.