README.md

metadata

language:
  - en
license: apache-2.0
tags:
  - generated_from_trainer
base_model: openai/whisper-small
datasets:
  - mozilla-foundation/common_voice_17_0
  - google/fleurs
  - facebook/voxpopuli
metrics:
  - wer
model-index:
  - name: Whisper Medium en
    results:
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: Common Voice 17.0
          type: mozilla-foundation/common_voice_17_0
          config: en
          split: test
          args: en
        metrics:
          - type: wer
            value: 13.5791343916967
            name: Wer
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: google/fleurs
          type: google/fleurs
          config: en_us
          split: test
        metrics:
          - type: wer
            value: 7.32
            name: WER
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: facebook/voxpopuli
          type: facebook/voxpopuli
          config: en
          split: test
        metrics:
          - type: wer
            value: 7.06
            name: WER
pipeline_tag: automatic-speech-recognition

Whisper small mixed-English

This model is a fine-tuned version of openai/whisper-small on the "en" datasets:

• mozilla-foundation/common_voice_17_0
• google/fleurs
• facebook/voxpopuli

It achieves the following results on the evaluation set:

• Loss: 0.3741
• Wer: 13.5791

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: 64
• eval_batch_size: 16
• 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: 5000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2018	0.2	1000	0.3925	14.4841
0.1553	0.4	2000	0.3887	14.0259
0.1545	0.6	3000	0.3805	14.0316
0.1223	0.8	4000	0.3776	13.6450
0.131	1.0	5000	0.3741	13.5791

Framework versions

• Transformers 4.42.0.dev0
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1