whisper-small-es / README.md
Drazcat's picture
update model card README.md
0a328c7
metadata
language:
  - es
license: apache-2.0
tags:
  - hf-asr-leaderboard
  - generated_from_trainer
datasets:
  - Drazcat/Cencosud
metrics:
  - wer
model-index:
  - name: Whisper Small Es - GoCloud
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: 30seg
          type: Drazcat/Cencosud
          args: 'config: es, split: test'
        metrics:
          - name: Wer
            type: wer
            value: 0

Whisper Small Es - GoCloud

This model is a fine-tuned version of openai/whisper-small on the 30seg dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0028
  • Wer: 0.0

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: 8
  • 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: 25
  • training_steps: 200
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
0.2944 5.56 50 0.1392 79.6117
0.08 11.11 100 0.0569 46.0472
0.0204 16.67 150 0.0086 0.0
0.0028 22.22 200 0.0028 0.0

Framework versions

  • Transformers 4.25.0.dev0
  • Pytorch 1.12.1+cu113
  • Datasets 2.7.1
  • Tokenizers 0.13.2