README.md

metadata

language:
  - ur
license: apache-2.0
tags:
  - automatic-speech-recognition
  - robust-speech-event
datasets:
  - mozilla-foundation/common_voice_8_0
metrics:
  - wer
  - cer
model-index:
  - name: wav2vec2-urdu-V8-Abid
    results:
      - task:
          type: automatic-speech-recognition
          name: Speech Recognition
        dataset:
          type: mozilla-foundation/common_voice_8_0
          name: Common Voice ur
          args: ur
        metrics:
          - type: wer
            value: 39.52
            name: Test WER
            args:
              - learning_rate: 0.00007
              - train_batch_size: 64
              - eval_batch_size: 8
              - seed: 42
              - gradient_accumulation_steps: 4
              - total_train_batch_size: 128
              - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
              - lr_scheduler_type: linear
              - lr_scheduler_warmup_steps: 100
              - num_epochs: 100
              - mixed_precision_training: Native AMP
          - type: cer
            value: 17.6
            name: Test CER
            args:
              - learning_rate: 0.00007
              - train_batch_size: 64
              - eval_batch_size: 8
              - seed: 42
              - gradient_accumulation_steps: 4
              - total_train_batch_size: 128
              - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
              - lr_scheduler_type: linear
              - lr_scheduler_warmup_steps: 100
              - num_epochs: 100
              - mixed_precision_training: Native AMP

wav2vec2-60-Urdu-V8

This model is a fine-tuned version of kingabzpro/wav2vec2-urdu on the common_voice dataset. It achieves the following results on the evaluation set:

• Loss: 4.9192
• Wer: 0.4741
• Cer: 0.2504

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 7e-05
• train_batch_size: 64
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 256
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
2.8836	16.62	50	4.7827	0.5011	0.2625
0.6992	33.31	100	3.5358	0.4882	0.2537
0.6321	49.92	150	4.9054	0.4774	0.2519
0.4669	66.62	200	5.9508	0.4719	0.2513
0.3119	83.31	250	5.5791	0.4745	0.2508
0.2788	99.92	300	4.9192	0.4741	0.2504

Framework versions

• Transformers 4.17.0.dev0
• Pytorch 1.10.2+cu102
• Datasets 1.18.2.dev0
• Tokenizers 0.11.0