raiyan007/whisper-tiny-common16.1

Whisper Tiny Bengali - Raiyan Ahmed

This model is a fine-tuned version of openai/whisper-tiny on the Common Voice 16.1 dataset. It achieves the following results on the evaluation set:

• Loss: 0.1982
• Wer: 49.3515

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: 5e-05
• train_batch_size: 16
• 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
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.5615	0.3021	200	0.5681	92.0473
0.4104	0.6042	400	0.4525	83.0059
0.336	0.9063	600	0.3315	74.3195
0.257	1.2085	800	0.3217	75.2095
0.2262	1.5106	1000	0.2550	65.7941
0.1906	1.8127	1200	0.2147	59.0769
0.1924	2.1148	1400	0.2816	67.6071
0.1886	2.4169	1600	0.2658	68.2982
0.175	2.7190	1800	0.2401	65.5598
0.1268	3.0211	2000	0.2279	57.7160
0.1206	3.3233	2200	0.2190	58.5680
0.1085	3.6254	2400	0.2048	54.5160
0.1049	3.9275	2600	0.1929	53.0769
0.047	4.2296	2800	0.2100	52.4805
0.0452	4.5317	3000	0.2054	50.8852
0.0411	4.8338	3200	0.1982	49.3515

Framework versions

• Transformers 4.41.0.dev0
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1