whisper-small-enhanced-hindi-10dB

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: 1.5528
• Wer: 57.6431

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
2.3087	0.61	50	1.9565	101.3315
1.3628	1.22	100	1.2862	83.4083
1.1319	1.83	150	1.0950	79.0334
0.9559	2.44	200	0.9573	74.3905
0.807	3.05	250	0.8252	71.1655
0.6268	3.66	300	0.6903	67.2488
0.5039	4.27	350	0.6466	64.4907
0.4738	4.88	400	0.6077	62.8566
0.3599	5.49	450	0.5964	60.7902
0.3225	6.1	500	0.6001	59.4761
0.2599	6.71	550	0.5930	58.5509
0.1658	7.32	600	0.6158	58.4731
0.1666	7.93	650	0.6172	58.0581
0.1032	8.54	700	0.6521	58.7152
0.081	9.15	750	0.6857	58.7930
0.0606	9.76	800	0.7020	57.9457
0.0345	10.37	850	0.7422	57.9284
0.0342	10.98	900	0.7622	57.5826
0.023	11.59	950	0.7787	57.8074
0.017	12.2	1000	0.8223	58.4299
0.0159	12.8	1050	0.8384	57.6604
0.0101	13.41	1100	0.8538	58.3607
0.012	14.02	1150	0.8634	57.8765
0.0092	14.63	1200	0.8762	57.5134
0.0077	15.24	1250	0.9077	58.6201
0.007	15.85	1300	0.9194	58.2310
0.006	16.46	1350	0.9194	57.1935
0.0051	17.07	1400	0.9427	57.4788
0.0044	17.68	1450	0.9613	57.5307
0.0037	18.29	1500	0.9750	57.3578
0.0038	18.9	1550	0.9620	57.1070
0.0037	19.51	1600	0.9793	57.2021
0.0028	20.12	1650	1.0002	57.6690
0.0023	20.73	1700	1.0171	57.0465
0.0023	21.34	1750	1.0344	56.4499
0.0024	21.95	1800	1.0231	56.9168
0.0017	22.56	1850	1.0420	56.6229
0.0016	23.17	1900	1.0599	57.6690
0.001	23.78	1950	1.0659	57.7641
0.0012	24.39	2000	1.0818	56.7093
0.001	25.0	2050	1.0874	57.0984
0.0008	25.61	2100	1.1034	57.5220
0.0006	26.22	2150	1.1275	56.7353
0.0004	26.83	2200	1.1528	57.1330
0.0002	27.44	2250	1.1668	56.5537
0.0001	28.05	2300	1.1935	56.6142
0.0001	28.66	2350	1.2282	56.3289
0.0001	29.27	2400	1.2547	56.7266
0.0001	29.88	2450	1.2814	56.4413
0.0001	30.49	2500	1.3142	56.8822
0.0	31.1	2550	1.3535	56.8995
0.0	31.71	2600	1.3759	57.0033
0.0	32.32	2650	1.4102	57.2454
0.0	32.93	2700	1.4299	56.8044
0.0	33.54	2750	1.4650	57.2886
0.0	34.15	2800	1.4906	57.3405
0.0	34.76	2850	1.5145	57.5739
0.0	35.37	2900	1.5377	57.5480
0.0	35.98	2950	1.5461	57.5480
0.0	36.59	3000	1.5528	57.6431

Framework versions

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