Adding preserve_empty_lines option #103
Conversation
|
This is going to need some more work. I guess all of my testing was with njobs=1. With parallel jobs, it seems the flattener doesn't the empty lines (at least with espeak). I'll look into fixing it. |
|
It turns out this was an existing bug in the library. With njobs>1 and an empty input (or one that gets preprocessed to empty), Even if you decide not to merge this new feature, it might still be a good idea to add this fix (or equivalent) |
…pty_lines_v2 # Conflicts: # phonemizer/phonemize.py
Codecov Report
@@ Coverage Diff @@
## master #103 +/- ##
=========================================
Coverage 100.00% 100.00%
=========================================
Files 23 23
Lines 1139 1152 +13
=========================================
+ Hits 1139 1152 +13
Continue to review full report at Codecov.
|
|
Alright, this is almost ready for merge. Could you try and add a test or two in order to make codecov happy again, and then we're go :) PS: don't forget to pull, i've merged the master onto your branch to fix some conflicts. |
|
Okay, great! I'll have a look at this soon. |
|
These extra tests should hit those missed lines. [Sorry, I don't have access to the festival backend on my local computer at the moment, but I can't imagine any reason why they wouldn't pass.] |
|
Alright, we're go! I'll merge that and (if I have the right permissions) upload it to pipy tomorrow! Great work, thanks for that @jncasey ! |
|
BTW: you might want to close the issues related to this PR. |
Adding the feature I requested in #95.
Not stripping out the empty lines was causing problems with the festival backend and preserve_punctuation, so I took the approach of stripping out the empty lines pre-phonemization and then reinserting them afterward.
I want to flag that I changed the conversion of the input text to a list from a generator to list comprehension here since I needed to run through the list a second time to preserve the empty lines, and I thought this made the code easier to read than making another generator. I'm assuming that this won't make a big difference on performance given how I think phonemizer is generally used.