abidanoaman/urdu-teacher-stage1-asr

Urdu ASR Teacher - Stage 1 (Base Model)

This is the base Urdu ASR model being used as our Stage 1 teacher.

Model Details

• Base Model: kingabzpro/wav2vec2-large-xls-r-300m-Urdu
• Language: Urdu (ur)
• Task: Automatic Speech Recognition
• Strategy: Using pre-trained model directly (no fine-tuning due to hardware limits)
• Expected WER: ~30-35%

Why This Approach?

The base model is already well-trained on Urdu ASR. When hardware resources are limited, it's valid to use the base model directly as the Stage 1 teacher and proceed to multi-task training (Stage 2).

Usage

This model will be used in Phase 4 for multi-task training (emotion + gender).

from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor

processor = Wav2Vec2Processor.from_pretrained("abidanoaman/urdu-teacher-stage1-asr")
model = Wav2Vec2ForCTC.from_pretrained("abidanoaman/urdu-teacher-stage1-asr")

Next Steps

Load this model in Phase 4 (Teacher Stage 2) for multi-task training.