DiaCTC — Wav2Vec2 (ClArTTS)
Fine-tuned Wav2Vec2 CTC model for Arabic diacritic restoration from speech + undiacritized transcript, trained on ClArTTS (Classical Arabic).
This checkpoint is used with DiaCTC constrained WFST decoding. Try the interactive demo: DiaCTC Space.
Results (ClArTTS training)
| Test set | WER ↓ | DER ↓ |
|---|---|---|
| ClArTTS | 11.21 | 3.53 |
| ArVoice | 39.89 | 12.04 |
Base encoder: jonatasgrosman/wav2vec2-large-xlsr-53-arabic.
Usage
from diactc.models import Wav2Vec2DiacritizationModel
model = Wav2Vec2DiacritizationModel(
"rufaelfekadu/diactc-wav2vec2-clartts",
device="cuda",
use_blank_token=True,
use_no_diac_token=True,
use_unk_diac_token=True,
)
text, rtf = model.diacritize(
undiacritized_text,
audio_path,
constrained=True,
method="wfst",
)
Citation
@article{marew2026constrained,
title = {Constrained CTC Decoding for Efficient Diacritic Restoration},
author = {Marew, Rufael and Keleg, Amr and Aldarmaki, Hanan},
journal = {arXiv preprint arXiv:TBD},
year = {2026}
}