steja/whisper-small-persian

Whisper small Persian

This model is a fine-tuned version of openai/whisper-small on the mozilla-foundation/common_voice_11_0 fa dataset. It achieves the following results on the evaluation set:

• Loss: 0.4924
• Wer: 32.8995

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-06
• train_batch_size: 8
• eval_batch_size: 16
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• total_eval_batch_size: 64
• 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
0.5533	1.56	500	0.7044	54.5499
0.3951	3.12	1000	0.5893	47.5210
0.3296	4.67	1500	0.5429	42.6451
0.2662	6.23	2000	0.5223	40.6644
0.2535	7.79	2500	0.5045	38.5304
0.224	9.35	3000	0.5002	36.8822
0.2204	10.9	3500	0.4967	35.3076
0.2024	12.46	4000	0.4951	34.9883
0.2099	14.02	4500	0.4921	34.9842
0.1836	15.58	5000	0.4924	34.8995

Framework versions

• Transformers 4.25.1
• Pytorch 1.13.0+cu117
• Datasets 2.7.1
• Tokenizers 0.13.2