kingabzpro/whisper-large-v3-urdu

Whisper large V3 Urdu ASR Model 🥇

This model is a fine-tuned version of openai/whisper-large-v3 on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0204
• Wer: 21.4712
• Cer: 7.1975

Quick Usage

from transformers import pipeline

transcriber = pipeline(
  "automatic-speech-recognition", 
  model="kingabzpro/whisper-large-v3-turbo-urdu"
)

transcriber.model.generation_config.forced_decoder_ids = None
transcriber.model.generation_config.language = "ur"

transcription = transcriber("audio2.mp3")
print(transcription)

{'text': 'دیکھیے پانی کب تک بہتا اور مچھلی کب تک تیرتی ہے'}

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 3e-05
• train_batch_size: 8
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• training_steps: 1500

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.0261	0.5089	300	0.0254	30.0224	10.3646
0.0211	1.0170	600	0.0226	25.8588	8.5780
0.0121	1.5259	900	0.0206	24.2158	7.9412
0.0093	2.0339	1200	0.0195	21.3032	7.2018
0.0043	2.5428	1500	0.0204	21.4712	7.1975

Framework versions

• Transformers 4.52.2
• Pytorch 2.7.1+cu126
• Datasets 3.4.1
• Tokenizers 0.21.2

---

Evaluation

Urdu ASR Evaluation on Common Voice 17.0 (Test Split).

Metric	Value	Description
WER	26.019%	Word Error Rate (lower is better)
CER	9.426%	Character Error Rate
BLEU	59.446%	BLEU Score (higher is better)
ChrF	82.902	Character n-gram F-score

👉 Review the testing script: Testing Whisper Large V3 Urdu