README.md

metadata

language:
  - yue
license: apache-2.0
tags:
  - hf-asr-leaderboard
  - generated_from_trainer
datasets:
  - mozilla-foundation/common_voice_11
metrics:
  - wer
model-index:
  - name: Whisper Base Yue
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Common Voice 11.0 yue
          type: mozilla-foundation/common_voice_11
          config: unclear
          split: None
          args: 'config: yue, split: train'
        metrics:
          - name: Wer
            type: wer
            value: 69.58637469586375

Whisper Base Yue

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

• Loss: 0.3671
• Wer: 69.5864

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: 2e-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: 200
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0998	2.78	500	0.3500	71.4517
0.0085	5.56	1000	0.3671	69.5864

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.0+cu116
• Datasets 2.7.1
• Tokenizers 0.13.2