metadata
language:
- ja
license: other
tags:
- whisper-event
- generated_from_trainer
datasets:
- Elite35P-Server/EliteVoiceProject
metrics:
- wer
model-index:
- name: Whisper Small Japanese Elite
results:
- task:
name: Automatic Speech Recognition
type: automatic-speech-recognition
dataset:
name: Elite35P-Server/EliteVoiceProject twitch
type: Elite35P-Server/EliteVoiceProject
config: twitch
split: test
args: twitch
metrics:
- name: Wer
type: wer
value: 23.296888141295206
Whisper Small Japanese Elite
This model is a fine-tuned version of openai/whisper-small on the Elite35P-Server/EliteVoiceProject twitch dataset. It achieves the following results on the evaluation set:
- Loss: 0.9180
- Wer: 23.2969
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: 16
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: constant_with_warmup
- lr_scheduler_warmup_steps: 100
- training_steps: 10000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.0033 | 18.0 | 1000 | 0.6728 | 25.3154 |
| 0.008 | 37.0 | 2000 | 0.6984 | 23.3810 |
| 0.0002 | 56.0 | 3000 | 0.7486 | 24.4743 |
| 0.0001 | 75.0 | 4000 | 0.7753 | 24.4743 |
| 0.0 | 94.0 | 5000 | 0.8014 | 24.0538 |
| 0.0 | 113.0 | 6000 | 0.8244 | 24.3902 |
| 0.0 | 132.0 | 7000 | 0.8468 | 23.8015 |
| 0.0 | 150.0 | 8000 | 0.8699 | 23.4651 |
| 0.0 | 169.0 | 9000 | 0.8936 | 23.2128 |
| 0.0 | 188.0 | 10000 | 0.9180 | 23.2969 |
Framework versions
- Transformers 4.26.0.dev0
- Pytorch 1.13.1+cu117
- Datasets 2.8.1.dev0
- Tokenizers 0.13.2