Add Word-Level Alignment

[ibevers]

Description

• Fixes a bug that was leading to incorrect alignments.
• Updates the output to give transcription, sentence, word, and character timestamps.
• Adds a recursive evaluation function for ensuring the differences in timestamps between two alignments are under a certain threshold.
• Adds a test that validates that the generated alignments match a publicly available alignment with a tolerance of .1 second for every start and end timestamp for every aligned level present in both alignments.
• Uses SenseLab native data types more (can be fully converted to SenseLab types in the future).
• Adds optional levels so the user can include or exclude utterance, word, and character-level alignments.
• Adds a tutorial for forced alignment.
• Adds a docs.md for forced alignment.

Related Issue(s)

#14

Motivation and Context

Corrects buggy forced alignment and adds more granular alignment (word and character), which is useful for various purposes. This is a character-based approach. I plan to add a phoneme-based approach in the future to compare to this.

How Has This Been Tested?

Added test to compare with publicly available alignment. Plan to do more comprehensive evaluation on TIMIT as soon as possible.

Screenshots (if appropriate):

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)

Checklist:

• I have added tests to cover my changes.
• All new and existing tests passed.
• My code follows the code style of this project.

[ibevers]

• Identify where the timestamps are calculated for indices
• Ensure indices for words are present at that point
• Update to calculate timestamps for words
• Ensure word timestamps are stored in final result

[ibevers]

Update 12/09:
Reviewed code and identified what needs to be changed. Still figuring out how exactly it should be changed.

Update 12/17:
The code works and returns an aligned ScriptLine for sentences, words, and characters. I am now fixing discrepancies with the previous tests. I also plan to add a correctness test for at least one labeled example.

[ibevers]

@fabiocat93 Would you kindly allow the workflow to run tests on this? All the forced alignment-relevant tests pass. I haven't run all the tests locally because I still have to address the squim issue.

[ibevers]

@fabiocat93 I addressed all your feedback:) Mergable?

[ibevers]

• Add links in docs.md

[fabiocat93]

Hi @ibevers , it looks good to me. You only need unit tests to pass

[ibevers]

Hi @ibevers , it looks good to me. You only need unit tests to pass

Thanks! I think there might be an issue with the CI configuration since the error here is common to another unrelated PR--see here: #321

[900miles]

I unfortunately didn't have time for a thorough review but the notebook uses an outdated form of loading audio from file so it gives an error. A couple of other comments but it seems to run okay otherwise.

[900miles]

Looks good!

[ibevers]

@fabiocat93 mergable ("Merging is blocked.")?

[fabiocat93]

@fabiocat93 mergable ("Merging is blocked.")?