Khmer G2P (NCKL respelling)

A small character-level Transformer (7.5M parameters) that converts Khmer words to their NCKL Khmer-script respelling, with - between syllables:

αž€αŸ’αžšαž”αžΈ β†’ αž€αŸ’αžš-αž”αžΈ

Trained on phonsobon/khmer-G2P. Best validation word accuracy: 83.37% (epoch 150).

Use

pip install khmer-g2p
khmer-g2p αž€αŸ’αžšαž”αžΈ αž€αž˜αŸ’αž–αž»αž‡αžΆ
from khmer_g2p import KhmerG2P
g2p = KhmerG2P()                 # downloads this repo
g2p("αž€αŸ’αžšαž”αžΈ")                     # 'αž€αŸ’αžš-αž”αžΈ'
g2p(["αžŸαž˜αŸ’αžŠαŸαž…", "αž€αž˜αŸ’αž–αž»αž‡αžΆ"])        # list in, list out

Files

  • model.safetensors β€” weights
  • config.json β€” model hyper-parameters and source/target vocabularies
  • history.json, training_curves.png β€” training metrics

training curves

Downloads last month
12
Safetensors
Model size
7.46M params
Tensor type
F32
Β·
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Dataset used to train phonsobon/khmer-g2p