metadata

language:
  - de
license: apache-2.0
tags:
  - sbb-asr
  - generated_from_trainer
datasets:
  - marccgrau/sbbdata_allSNR
metrics:
  - wer
base_model: openai/whisper-small
model-index:
  - name: Whisper Large-v2 German SBB ASR
    results:
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: SBB Dataset 05.01.2023
          type: marccgrau/sbbdata_allSNR
          args: 'config: German, split: train, test, val'
        metrics:
          - type: wer
            value: 0.023462270133164237
            name: Wer

Whisper Large-v2 German SBB ASR

This model is a fine-tuned version of openai/whisper-small on the SBB Dataset 05.01.2023 dataset. It achieves the following results on the evaluation set:

Loss: 0.0277
Wer: 0.0235

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: 32
eval_batch_size: 32
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 100
training_steps: 600
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.3449	0.36	100	0.2158	0.0387
0.0647	0.71	200	0.0266	0.0197
0.0308	1.07	300	0.0315	0.0216
0.0188	1.42	400	0.0286	0.0197
0.0136	1.78	500	0.0298	0.0209
0.0089	2.14	600	0.0277	0.0235

Framework versions

Transformers 4.25.1
Pytorch 1.13.1
Datasets 2.8.0
Tokenizers 0.12.1