whisper-small-indian_eng

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

• Loss: 0.3169
• Wer: 10.6061

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: 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: 100
• training_steps: 3000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.5504	1.5873	100	0.4581	10.6061
0.058	3.1746	200	0.2114	8.8154
0.0126	4.7619	300	0.2365	8.1267
0.0078	6.3492	400	0.2513	7.9890
0.0026	7.9365	500	0.2596	7.1625
0.0027	9.5238	600	0.2676	8.1267
0.0031	11.1111	700	0.2842	7.3003
0.0013	12.6984	800	0.2707	18.7328
0.0004	14.2857	900	0.2814	13.6364
0.0006	15.8730	1000	0.2806	11.1570
0.0009	17.4603	1100	0.2906	11.0193
0.0004	19.0476	1200	0.2948	11.1570
0.0003	20.6349	1300	0.2984	10.8815
0.0002	22.2222	1400	0.3008	10.8815
0.0003	23.8095	1500	0.2992	10.7438
0.0001	25.3968	1600	0.3040	10.8815
0.0002	26.9841	1700	0.3088	10.7438
0.0001	28.5714	1800	0.3077	10.6061
0.0001	30.1587	1900	0.3098	10.4683
0.0001	31.7460	2000	0.3111	10.4683
0.0001	33.3333	2100	0.3119	10.4683
0.0001	34.9206	2200	0.3129	10.4683
0.0001	36.5079	2300	0.3142	10.4683
0.0001	38.0952	2400	0.3146	10.4683
0.0001	39.6825	2500	0.3153	10.6061
0.0001	41.2698	2600	0.3158	10.6061
0.0001	42.8571	2700	0.3162	10.6061
0.0001	44.4444	2800	0.3167	10.6061
0.0001	46.0317	2900	0.3169	10.6061
0.0001	47.6190	3000	0.3169	10.6061

Framework versions

• Transformers 4.42.0.dev0
• Pytorch 2.0.0
• Datasets 2.16.1
• Tokenizers 0.19.1