MMS-300m — Yemeni Sub-Dialect Identifier (AYDID)
MMS-300m fine-tuned on the AYDID corpus for 7-class Yemeni Arabic sub-dialect identification: Adeni, Badawi, Hadrami, Sana'ani, Ta'izzi, Tihami, and Standard Yemeni.
- Base model:
badrex/mms-300m-arabic-dialect-identifier - Dataset:
mansoorSaleh/AYDID - Task: audio classification (dialect ID)
Result
On the balanced AYDID test set (375 utterances/class), fine-tuning raises weighted F1 from a zero-shot 11.45% to 80.44% (+68.99 points).
Usage
import torch, librosa
from transformers import AutoFeatureExtractor, AutoModelForAudioClassification
model_id = "mansoorSaleh/mms-300m-aydid-did"
fe = AutoFeatureExtractor.from_pretrained(model_id)
model = AutoModelForAudioClassification.from_pretrained(model_id).eval()
wav, _ = librosa.load("clip.wav", sr=16000)
inputs = fe(wav, sampling_rate=16000, return_tensors="pt")
with torch.no_grad():
logits = model(**inputs).logits
pred = logits.argmax(-1).item()
print(model.config.id2label[pred]) # e.g. YEM_HA
Labels
YEM_AD Adeni · YEM_BA Badawi · YEM_HA Hadrami · YEM_SA Sana'ani ·
YEM_ST Standard · YEM_TA Ta'izzi · YEM_TI Tihami
Citation
@inproceedings{aydid2027,
title = {AYDID: A Sub-Dialectal Yemeni Arabic Corpus for Dialect
Identification and Speech Recognition},
author = {Ba Mahel, Mansoor S. M. and Wei, Jianguo and Yue, Xianghu and
Awn, Norah Saeed and Bamahel, Abdulaziz S.},
booktitle = {Proc. IEEE ICASSP},
year = {2027}
}
License: CC BY-NC-SA 4.0 (research use).