common_voice / README.md
Tomas1234's picture
Upload README.md
9e467de
metadata
language:
  - lt
license: apache-2.0
tags:
  - hf-asr-leaderboard
  - generated_from_trainer
datasets:
  - mozilla-foundation/common_voice_11_0
metrics:
  - wer
model-index:
  - name: Whisper Small lt - Lithuanian
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Common Voice 11.0
          type: mozilla-foundation/common_voice_11_0
          config: None
          split: None
          args: 'config: lt, split: test'
        metrics:
          - name: Wer
            type: wer
            value: 32.49711764004294

Whisper Small lt - Lithuanian

This model is a fine-tuned version of openai/whisper-small on the Common Voice 11.0 dataset. It achieves the following results on the evaluation set:

  • Loss: 0.3840
  • Wer: 32.4971

Model description

More information needed

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: 16
  • eval_batch_size: 8
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 250
  • training_steps: 4000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
0.3788 0.9 500 0.4432 45.1716
0.2087 1.8 1000 0.3671 37.6456
0.0961 2.7 1500 0.3548 35.5703
0.0479 3.6 2000 0.3609 34.1709
0.0157 4.5 2500 0.3665 33.3400
0.0089 5.4 3000 0.3775 32.7754
0.0038 6.29 3500 0.3826 32.5607
0.0033 7.19 4000 0.3840 32.4971

Framework versions

  • Transformers 4.26.0.dev0
  • Pytorch 1.12.1+cu113
  • Datasets 2.7.1
  • Tokenizers 0.13.2