deepdml/whisper-medium-mix-pt

Whisper Medium Mixed-Portuguese

This model is a fine-tuned version of openai/whisper-medium on the pt datasets:

• mozilla-foundation/common_voice_17_0
• google/fleurs
• facebook/multilingual_librispeech

It achieves the following results on the evaluation set:

• Loss: 0.1353
• Wer: 7.1230

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: 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: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1116	0.2	1000	0.1570	8.5824
0.105	0.4	2000	0.1484	7.9398
0.0783	0.6	3000	0.1374	7.4475
0.1703	0.8	4000	0.1370	7.2413
0.0977	1.0622	5000	0.1353	7.1230

Framework versions

• Transformers 4.42.0.dev0
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1