olaverse/diacnet-yor-viterbi

DiacNetYor-Viterbi

DiacNetYor-Viterbi is a lightweight, statistical full tonal and dot-below diacritization model for Yoruba (yo) text. It uses a bigram Viterbi decoder over a word-level candidate map.

Model Details

• Model Type: Word-level Bigram Viterbi Decoder
• File Size: 7.17 MB (yoruba_diacritizer.json)
• Supported Languages: Yoruba (yo)
• Metrics:
  • Word Accuracy: 71.66% (evaluated on 54,815 words)
• Dependencies: None (pure Python / zero dependencies)

Usage

Loaded and used via the unified olaverse SDK wrapper:

from olaverse.nlp.diacritizer import Diacritizer

diacritizer = Diacritizer(model="diacnet-yor-viterbi")
text = "Ojo lo si oja lana"
print(diacritizer.restore(text))
# Output: "Ọjọ́ ló sí ọjà lànà"

Links

• GitHub
• Olaverse Library Docs