whisper-small-enhanced-hindi

This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:

• Loss: 2.3627
• Wer: 95.1323

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: 64
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 3000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.0111	0.61	50	2.6830	173.1800
2.1003	1.22	100	2.0362	114.2054
1.884	1.83	150	1.8518	196.2390
1.7085	2.44	200	1.7161	104.2625
1.5726	3.05	250	1.5840	99.8011
1.3908	3.66	300	1.4421	98.4524
1.2463	4.27	350	1.3700	95.3484
1.2047	4.88	400	1.3240	97.6050
1.069	5.49	450	1.2952	96.8442
1.0265	6.1	500	1.2916	95.0026
0.9474	6.71	550	1.2525	94.6222
0.8052	7.32	600	1.2687	94.3109
0.7971	7.93	650	1.2595	92.5990
0.6551	8.54	700	1.2944	93.0140
0.6156	9.15	750	1.3545	93.7749
0.5445	9.76	800	1.3548	93.5328
0.4183	10.37	850	1.4263	93.6798
0.4311	10.98	900	1.4321	94.1726
0.3244	11.59	950	1.4985	94.0169
0.2604	12.2	1000	1.6052	93.9997
0.2403	12.8	1050	1.6366	94.4665
0.1669	13.41	1100	1.7287	94.4752
0.1648	14.02	1150	1.7606	96.0661
0.1184	14.63	1200	1.8454	94.9507
0.086	15.24	1250	1.9234	94.8037
0.0845	15.85	1300	1.9631	94.7000
0.0504	16.46	1350	2.0461	94.1121
0.0525	17.07	1400	2.1092	93.9478
0.0381	17.68	1450	2.1563	95.0718
0.0261	18.29	1500	2.2302	94.3109
0.0282	18.9	1550	2.2574	94.0775
0.0209	19.51	1600	2.2939	94.7432
0.0187	20.12	1650	2.3627	95.1323

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 1.12.1
• Datasets 2.16.1
• Tokenizers 0.15.0