whisper-small-tr / README.md
sgangireddy's picture
Update README.md
737012d
metadata
language:
  - tr
license: apache-2.0
tags:
  - whisper-event
  - generated_from_trainer
datasets:
  - mozilla-foundation/common_voice_11_0
metrics:
  - wer
model-index:
  - name: Whisper Small Turkish
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: mozilla-foundation/common_voice_11_0 tr
          type: mozilla-foundation/common_voice_11_0
          config: tr
          split: test
          args: tr
        metrics:
          - name: Wer
            type: wer
            value: 17.210750075918614

Whisper Small Turkish

This model is a fine-tuned version of openai/whisper-small on the mozilla-foundation/common_voice_11_0 tr dataset. It achieves the following results on the evaluation set:

  • Loss: 0.2611
  • Wer: 17.2108

Model description

The Whisper small model is fine-tuned on CommonVoice Turkish train+validation partitions. The model was fine-tuend for 1000 steps. Without fine-tuning the WER on test data was 23.7 (using whsiper-small and the CommonVoice 9 test data).

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 1e-05
  • train_batch_size: 64
  • eval_batch_size: 32
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 500
  • training_steps: 1000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
0.0693 3.05 1000 0.2611 17.2108

Framework versions

  • Transformers 4.26.0.dev0
  • Pytorch 1.13.0+cu117
  • Datasets 2.7.1.dev0
  • Tokenizers 0.13.2