metadata
language:
- toi
license: apache-2.0
tags:
- hf-asr-leaderboard
- generated_from_trainer
datasets:
- zambezi_voice/tonga
metrics:
- wer
model-index:
- name: Whisper Medium - Claytone Aikasote
results:
- task:
name: Automatic Speech Recognition
type: automatic-speech-recognition
dataset:
name: tonga
type: zambezi_voice/tonga
config: null
split: None
args: 'config: toi, split: dev'
metrics:
- name: Wer
type: wer
value: 59.61628451099672
Whisper Medium - Claytone Aikasote
This model is a fine-tuned version of openai/whisper-medium on the tonga dataset. It achieves the following results on the evaluation set:
- Loss: 2.8215
- Wer: 59.6163
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: 4
- eval_batch_size: 4
- 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: 4000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.4727 | 1.47 | 500 | 2.0656 | 70.8002 |
| 0.2033 | 2.95 | 1000 | 2.0971 | 67.6416 |
| 0.0658 | 4.42 | 1500 | 2.3894 | 62.0262 |
| 0.0281 | 5.9 | 2000 | 2.5443 | 62.2134 |
| 0.0104 | 7.37 | 2500 | 2.6873 | 61.8390 |
| 0.0046 | 8.85 | 3000 | 2.7252 | 60.6458 |
| 0.0004 | 10.32 | 3500 | 2.7891 | 60.8563 |
| 0.0003 | 11.8 | 4000 | 2.8215 | 59.6163 |
Framework versions
- Transformers 4.26.0.dev0
- Pytorch 1.12.1+cu113
- Datasets 2.7.1
- Tokenizers 0.13.2