Scrya/whisper-medium-ms-augmented

Whisper Medium MS - Augmented

This model is a fine-tuned version of openai/whisper-medium on the google/fleurs dataset. It achieves the following results on the evaluation set:

• Loss: 0.2066
• Wer: 9.5784
• Cer: 2.8109

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

Training:

• google/fleurs (train+validation)

Evaluation:

• google/fleurs (test)

Training procedure

Datasets were augmented on-the-fly using audiomentations via PitchShift and TimeStretch transformations at p=0.3.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-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: linear
• lr_scheduler_warmup_steps: 100
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.0876	2.15	200	0.1949	10.3105	3.0685
0.0064	4.3	400	0.1974	9.7004	2.9596
0.0014	6.45	600	0.2031	9.6190	2.8955
0.001	8.6	800	0.2058	9.6055	2.8440
0.0009	10.75	1000	0.2066	9.5784	2.8109

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.0+cu117
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2