lab2-whisper-sv / README.md
SodraZatre's picture
Update README.md
8e575c5
|
raw
history blame
No virus
2.29 kB
metadata
language:
  - hi
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 Hi - Swedish
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Common Voice 11.0
          type: mozilla-foundation/common_voice_11_0
          config: config.json
          split: None
          args: 'config: hi, split: test'
        metrics:
          - name: Wer
            type: wer
            value: 19.894598155467722

Whisper Small Hi - Swedish

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: 0.3260
  • Wer: 19.8946

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: 500
  • training_steps: 4000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
0.3144 0.65 500 0.3244 24.0623
0.1321 1.29 1000 0.2977 21.5563
0.1318 1.94 1500 0.2788 20.9190
0.052 2.59 2000 0.2852 20.3329
0.0203 3.23 2500 0.3017 19.8677
0.0174 3.88 3000 0.3008 19.9941
0.0083 4.53 3500 0.3216 20.0022
0.0039 5.17 4000 0.3260 19.8946

Framework versions

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