aloudreader/uk-stress-token-classifier-ct2-int8

uk-stress-token-classifier, CTranslate2 INT8

aloudreader/uk-stress-token-classifier (tok-v5: XLM-R-uk classifier trained on stress heard in audiobooks) converted to CTranslate2 with INT8 weights.

	PyTorch FP32	CT2 INT8
size	420 MB	106 MB
CPU latency per target	31 ms	5.2 ms

The head that scores candidate stress positions is in head.npz (and head.pt); coverage.json lists the forms the classifier may decide (seen at least 80 times in training), as in the original.

Used by the light mode of the accentuk package (with spaCy uk_core_news_sm for morphology). Measured on 2026-10-13, word accuracy:

test	service / full mode	light mode
200 most frequent ambiguous forms of Common Voice uk, stress heard in the recordings	95.88%	95.54%
modern-text gold, 3,887 words	98.07%	97.63%
lang-uk stress benchmark, words / heteronyms	96.10% / 84.62%	96.10% / 84.62%