fixes to --preserve-punctuation

[jncasey]

This reworks a few things related to preserving punctuation.

• Fixes the inconsistency described in Reworking --preserve_punctuation and --strip #106. Now word separators appear in the same locations whether preserve_punctuation is True or False.
• Addresses part of the problem described in The punctuation mark is not separated correctly when using space as the phone separator in v0.3.1 #104 (but doesn't return the output to what it was in 0.3, which I believe was incorrect – they'd also need to use a non-None word separator, maybe, to get what they're after?)
• Fixes Maximum recursion depth exceeded with --preserve-punctuation and long documents #108 by refactoring the restore method to be iterative instead of recursive

A number of the tests had to be updated due to that first bullet point.

[codecov]

Codecov Report

Merging #119 (d2e5726) into master (d9315b9) will not change coverage.
The diff coverage is 100.00%.

@@            Coverage Diff            @@
##            master      #119   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           23        23           
  Lines         1152      1168   +16     
=========================================
+ Hits          1152      1168   +16     

Impacted Files	Coverage Δ
phonemizer/backend/base.py	100.00% <100.00%> (ø)
phonemizer/backend/espeak/espeak.py	100.00% <100.00%> (ø)
phonemizer/punctuation.py	100.00% <100.00%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d9315b9...d2e5726. Read the comment docs.

[jncasey]

@hadware I think this is good to go, but it definitely changes the output when preserving punctuation compared to previous versions of phonemizer. This way seems much more consistent to me – the output will have the same number of word separators whether punctuation is being preserved or not.

See the tests I modified for examples of the new output.

[hadware]

Alright, super cool.

I'll review the PR probably tomorrow, and hopefully we can merge this. This is fantastic, thanks again for your work!

[jncasey]

Great! Let me know if you need me to make any changes.

And thank you for this project. It's replaced my DIY solution of a transformers model trained on the CMUDict, and is so much faster and more accurate.

[hadware]

This is looking good! I'm merging this. I'd like to add a test with the large e-book from #108 , then i'll release a new version (Probably 3.2.1 since this contains some pretty major changes)

[jncasey]

Great! This also closed #106, and makes PR #112 redundant

[hadware]

Oh, btw. Since you seem to have a very good understanding of the whole lib, would you mind if I ask for your opinion on future PR's (from other github users)? I'm also making you "triager" for this repo.

[jncasey]

Happy to help as much as I can!