whisper-small-clean-hi

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: 0.5136
• Wer: 28.2379

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: 48
• eval_batch_size: 24
• 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
1.5251	0.46	50	1.2276	88.8034
0.7311	0.92	100	0.6706	50.3372
0.5582	1.38	150	0.5367	43.6798
0.4555	1.83	200	0.4448	43.1783
0.3326	2.29	250	0.3594	36.2182
0.2394	2.75	300	0.2507	33.5380
0.1449	3.21	350	0.2294	32.7252
0.1407	3.67	400	0.2144	30.6070
0.1048	4.13	450	0.2125	29.6299
0.0854	4.59	500	0.2085	29.1371
0.0762	5.05	550	0.2125	28.4109
0.0445	5.5	600	0.2168	28.4973
0.0474	5.96	650	0.2197	28.2725
0.0249	6.42	700	0.2324	28.2898
0.0267	6.88	750	0.2287	27.2696
0.0144	7.34	800	0.2440	27.2869
0.0154	7.8	850	0.2524	27.3733
0.008	8.26	900	0.2648	27.1312
0.0103	8.72	950	0.2602	27.9353
0.0066	9.17	1000	0.2718	28.3330
0.0073	9.63	1050	0.2705	27.4771
0.0053	10.09	1100	0.2828	27.5030
0.0044	10.55	1150	0.2882	27.2004
0.0045	11.01	1200	0.2892	27.5117
0.0037	11.47	1250	0.2961	27.3215
0.0031	11.93	1300	0.2934	27.0534
0.0022	12.39	1350	0.3014	27.1053
0.003	12.84	1400	0.3077	26.5779
0.0022	13.3	1450	0.3096	26.8373
0.002	13.76	1500	0.3123	26.5347
0.0017	14.22	1550	0.3186	26.8632
0.0016	14.68	1600	0.3255	26.6903
0.0012	15.14	1650	0.3329	26.4396
0.0015	15.6	1700	0.3336	27.0188
0.0009	16.06	1750	0.3361	26.4569
0.001	16.51	1800	0.3483	26.4655
0.0014	16.97	1850	0.3533	26.2666
0.0004	17.43	1900	0.3581	26.0678
0.0004	17.89	1950	0.3688	26.5087
0.0003	18.35	2000	0.3738	26.2148
0.0004	18.81	2050	0.3729	26.1197
0.0005	19.27	2100	0.3850	25.8776
0.0002	19.72	2150	0.3874	25.9900
0.0004	20.18	2200	0.3927	25.9727
0.0	20.64	2250	0.4037	25.9381
0.0	21.1	2300	0.4133	25.9208
0.0001	21.56	2350	0.4188	25.5836
0.0	22.02	2400	0.4266	25.8776
0.0	22.48	2450	0.4380	26.1715
0.0	22.94	2500	0.4473	25.6268
0.0	23.39	2550	0.4604	26.0418
0.0	23.85	2600	0.4681	26.1802
0.0	24.31	2650	0.4833	26.1197
0.0	24.77	2700	0.4883	26.2234
0.0	25.23	2750	0.4993	26.4914
0.0	25.69	2800	0.5031	26.7768
0.0	26.15	2850	0.5077	26.6211
0.0	26.61	2900	0.5102	27.1658
0.0	27.06	2950	0.5123	28.1688
0.0	27.52	3000	0.5136	28.2379

Framework versions

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