sharjeel103/whisper-tiny-urdu

Whisper Tiny Urdu

This model is a fine-tuned version of openai/whisper-tiny on the Common Voice 17.0 dataset.

👉 Review the testing script: Testing Urdu Whisper tiny

Model description

Whisper Tiny Urdu ASR Model This Whisper Tiny model has been fine-tuned on the Common Voice 17 dataset, which includes over 55 hours of Urdu speech data. The model was trained twice with different hyperparameters to optimize its performance:

Despite being the smallest variant in its family, this model achieves Good performance for Urdu ASR tasks. It can be used for deployment on small devices, offering an excellent balance between efficiency and accuracy.

Note: The test split was included during training. Therefore, any metrics previously reported on this split do not reflect real-world generalization and have been removed to avoid confusion.

Intended uses & limitations

This model is particularly suited for applications on edge devices with limited computational resources. Additionally, it can be converted to a FasterWhisper model using the CTranslate2 library, allowing for even faster inference on devices with lower processing power.

Evaluation

Urdu ASR Evaluation on urdu-asr/csalt-voice (Validation Split).

Metric	Value	Description
WER	64.961%	Word Error Rate (lower is better)
CER	42.488%	Character Error Rate
BLEU	16.710%	BLEU Score (higher is better)
ChrF	43.545	Character n-gram F-score

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 4e-05
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• training_steps: 3000
• mixed_precision_training: Native AMP

Framework versions

• Transformers 4.42.3
• Pytorch 2.1.2
• Datasets 2.20.0
• Tokenizers 0.19.1