whisper-tiny-sv

This model is a fine-tuned version of openai/whisper-tiny on the dataset/riksdagen audiofolder dataset. It achieves the following results on the evaluation set:

• Loss: 0.6435
• Wer: 0.3701

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: 32
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• total_train_batch_size: 128
• total_eval_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• training_steps: 5000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.0032	0.08	250	1.0075	0.5063
0.8983	0.17	500	0.8945	0.4649
0.8227	0.25	750	0.8336	0.4491
0.777	0.33	1000	0.7931	0.4314
0.7728	0.42	1250	0.7640	0.4217
0.7141	0.5	1500	0.7407	0.4134
0.7208	0.58	1750	0.7225	0.4023
0.6911	0.66	2000	0.7083	0.3942
0.6924	0.75	2250	0.6948	0.3911
0.6702	0.83	2500	0.6849	0.3884
0.663	0.91	2750	0.6766	0.3769
0.6548	1.0	3000	0.6686	0.3759
0.638	1.08	3250	0.6627	0.3728
0.6222	1.16	3500	0.6574	0.3733
0.6323	1.25	3750	0.6528	0.3691
0.6192	1.33	4000	0.6498	0.3688
0.633	1.41	4250	0.6469	0.3677
0.6229	1.5	4500	0.6451	0.3681
0.6246	1.58	4750	0.6439	0.3706
0.6214	1.66	5000	0.6435	0.3701

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.12.0a0+8a1a93a
• Datasets 2.7.1
• Tokenizers 0.13.2