v0.2.2

First public release of nyra-forced-aligner: precise word and phone timestamps for conversational speech, including fillers, cut-offs and vocal sounds.

pip install nyra-forced-aligner

from nyra_align import Aligner
result = Aligner().align("interview.wav", "so [UM] I th- I think we should go")

• English model nyralabs/nyra_forced_aligner_en, downloaded automatically on first use
• 19.9 ms mean word-boundary error on Buckeye (full test set)
• Long recordings in one call; GPU (fp16) and CPU
• Works directly on CrisperWhisper 2.0 verbatim transcripts

PyPI: https://pypi.org/project/nyra-forced-aligner/0.2.2/ · Blog: https://nyra-labs.com/research/nyra-forced-aligner