metadata
license: apache-2.0
base_model: openai/whisper-small
tags:
- generated_from_trainer
datasets:
- librispeech-clean
metrics:
- wer
model-index:
- name: Whisper Small English 1h
results:
- task:
name: Automatic Speech Recognition
type: automatic-speech-recognition
dataset:
name: Librispeech
type: librispeech-clean
config: default
split: None
args: 'config: english, split: test'
metrics:
- name: Wer
type: wer
value: 53.45675203126608
Whisper Small English 1h
This model is a fine-tuned version of openai/whisper-small on the Librispeech dataset. It achieves the following results on the evaluation set:
- Loss: 1.8110
- Wer: 53.4568
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-07
- 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: 1000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.0582 | 10.0 | 200 | 1.8847 | 56.4620 |
| 0.0495 | 20.0 | 400 | 1.8598 | 55.1579 |
| 0.042 | 30.0 | 600 | 1.8303 | 54.2240 |
| 0.0309 | 40.0 | 800 | 1.8152 | 53.7118 |
| 0.0323 | 50.0 | 1000 | 1.8110 | 53.4568 |
Framework versions
- Transformers 4.40.1
- Pytorch 2.3.0+cu121
- Datasets 2.19.0
- Tokenizers 0.19.1