fstalign-1.5.0

Improve support for non-English languages and non-ascii characters

What's Changed

• Fix for when empty strings are in the symbol table by @qmac in #23
• Add nlp prepunctuation column by @ajhinsvark in #21
• Better handling of case differences for non-ascii inputs by @qmac in #24

Full Changelog: 1.4.0...1.5.0

fstalign-1.4.1

Full Changelog: 1.4.0...1.4.1

• Bugfix: Fix for when empty strings are in the symbol table

fstalign-1.4.0

What's Changed

• Adding scripts to easily gather runtime metrics on new releases by @qmac in #16
• FSTALIGN-42: Add more synonyms by @pique0822 in #17
• Shuffling test files around by @qmac in #18
• FSTALIGN-42: [BUG] Removing duplicate entries by @pique0822 in #19
• Ignore hyphenated compound words for WER measurements by @qmac in #20
• Handling crashes using new Levenshtein approach by @qmac in #22

New Contributors

• @pique0822 made their first contribution in #17

Full Changelog: 1.3.0...1.4.0

fstalign-1.3.0

This release includes a new first-pass approximation of the alignment to speedup the real alignment process. The utility will consume more ram than before, but should be much faster in general. The alignment might vary a little from previous alignments.

To disable this pre-processing, once can pass --disable-approx-alignment on the command line.

https://github.com/revdotcom/fstalign/wiki/fstalign-1.3.0-resource-benchmarking

fstalign-1.2.0

This release contains:

• Make adapted composition the default - performance should be ~30% faster on most files
• Add flag to gather precision and recall metrics on casing

fstalign-1.1.0

This release contains:

• Removing kaldi as a dependency for FST hypothesis handling (README reflects new build instructions)
• Minor documentation changes

Initial public release

Migrating for Github packages to Docker Hub for container registry (#5)

* Using docker hub and latest build-push-action version

* Pushing to latest tag too

* Multiple target platforms

* Nevermind, can't support arm

* Updated readme