README.md

metadata

license: apache-2.0
tags:
  - generated_from_trainer
datasets:
  - voxpopuli
metrics:
  - wer
base_model: openai/whisper-large-v2
model-index:
  - name: whisper-large-v2-german
    results:
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: voxpopuli
          type: voxpopuli
          config: de
          split: test
          args: de
        metrics:
          - type: wer
            value: 0.12201852946974177
            name: Wer

whisper-large-v2-german

This model is a fine-tuned version of openai/whisper-large-v2 on the voxpopuli dataset. It achieves the following results on the evaluation set:

• Loss: 0.2841
• Wer Ortho: 0.1517
• Wer: 0.1220

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: 3e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 50
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
0.2616	1.0	1679	0.2695	0.1601	0.1303
0.1801	2.0	3358	0.2690	0.1554	0.1235
0.1185	3.0	5037	0.2841	0.1517	0.1220

Framework versions

• Transformers 4.30.0.dev0
• Pytorch 2.0.1+cu117
• Datasets 2.13.1
• Tokenizers 0.13.3