cawoylel/windanam_whisper-medium

windanam-whisper-medium

This model is a fine-tuned version of openai/whisper-medium on the cawoylel/FulaSpeechCorpora-splited-noise_augmented dataset. The finetuning was done on the train and test splits of the dataset. It achieves the following results on the evaluation set:

• Loss: 0.1407
• Wer: 0.2006

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: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• total_train_batch_size: 32
• total_eval_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.5665	0.16	1000	0.3283	0.3337
0.3998	0.31	2000	0.2489	0.2825
0.35	0.47	3000	0.2061	0.2549
0.3084	0.62	4000	0.1842	0.2263
0.2603	0.78	5000	0.1693	0.2169
0.2414	0.93	6000	0.1592	0.2097
0.1604	1.09	7000	0.1519	0.2009
0.1584	1.24	8000	0.1474	0.2007
0.1442	1.4	9000	0.1427	0.1980
0.1391	1.55	10000	0.1407	0.2006

Framework versions

• Transformers 4.35.0.dev0
• Pytorch 2.1.0+cu121
• Datasets 2.14.6
• Tokenizers 0.14.1