renikud

Hebrew grapheme-to-phoneme (G2P) model. Converts unvocalized Hebrew text to IPA phonemes.

• 85.1% word accuracy on heb-g2p-benchmark
• Single .onnx file, no extra dependencies
• Punctuation-conditional: preserves punctuation when present

Usage

from renikud_onnx import G2P

g2p = G2P("model.onnx")
print(g2p.phonemize("שלום עולם"))
# → ʃalˈom ʔolˈam

Training

See thewh1teagle/renikud-v4 for the full training code.