whisper-small-Fleurs_AMMI_AFRIVOICE_LRSC-ln-109hrs-v1

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: 0.9074
• Wer: 0.2063
• Cer: 0.0793

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: 4
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Use adamw_hf with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.4677	1.0	2880	0.7253	0.6948	0.3565
0.6293	2.0	5760	0.5281	0.5550	0.3003
0.4895	3.0	8640	0.4502	0.4045	0.2119
0.3993	4.0	11520	0.4109	0.4834	0.2918
0.3241	5.0	14400	0.3989	0.4870	0.2655
0.2551	6.0	17280	0.4082	0.4294	0.2432
0.1918	7.0	20160	0.4232	0.3710	0.1913
0.1369	8.0	23040	0.4381	0.4223	0.2191
0.0967	9.0	25920	0.4774	0.3570	0.1636
0.0709	10.0	28800	0.4869	0.3268	0.1283
0.0544	11.0	31680	0.5251	0.2836	0.1034
0.04	12.0	34560	0.5354	0.2856	0.1043
0.0309	13.0	37440	0.5464	0.2683	0.0985
0.0247	14.0	40320	0.5621	0.2640	0.0933
0.021	15.0	43200	0.5743	0.2510	0.0862
0.0185	16.0	46080	0.5972	0.2540	0.0893
0.0161	17.0	48960	0.5969	0.2399	0.0820
0.0143	18.0	51840	0.6073	0.2394	0.0805
0.013	19.0	54720	0.6256	0.2350	0.0814
0.0117	20.0	57600	0.6198	0.2298	0.0800
0.0103	21.0	60480	0.6410	0.2330	0.0830
0.0097	22.0	63360	0.6626	0.2373	0.0821
0.009	23.0	66240	0.6666	0.2279	0.0770
0.0083	24.0	69120	0.6686	0.2269	0.0779
0.0075	25.0	72000	0.6851	0.2220	0.0793
0.0071	26.0	74880	0.6944	0.2261	0.0784
0.0065	27.0	77760	0.7010	0.2240	0.0793
0.0058	28.0	80640	0.6999	0.2282	0.0812
0.0055	29.0	83520	0.7135	0.2208	0.0778
0.0054	30.0	86400	0.7268	0.2240	0.0806
0.0047	31.0	89280	0.7210	0.2208	0.0774
0.0045	32.0	92160	0.7255	0.2180	0.0776
0.0045	33.0	95040	0.7512	0.2179	0.0776
0.0037	34.0	97920	0.7662	0.2177	0.0812
0.0038	35.0	100800	0.7562	0.2087	0.0757
0.0036	36.0	103680	0.7524	0.2093	0.0740
0.0033	37.0	106560	0.7719	0.2136	0.0783
0.003	38.0	109440	0.7792	0.2149	0.0764
0.0029	39.0	112320	0.7883	0.2135	0.0761
0.0029	40.0	115200	0.7960	0.2098	0.0750
0.0026	41.0	118080	0.7779	0.2072	0.0750
0.0026	42.0	120960	0.7881	0.2155	0.0784
0.0024	43.0	123840	0.7936	0.2076	0.0740
0.0022	44.0	126720	0.8039	0.2071	0.0735
0.0022	45.0	129600	0.8125	0.2055	0.0740
0.0021	46.0	132480	0.8248	0.2118	0.0749
0.0019	47.0	135360	0.8211	0.2119	0.0777
0.0017	48.0	138240	0.8212	0.2100	0.0783
0.0019	49.0	141120	0.8303	0.2052	0.0744
0.0018	50.0	144000	0.8337	0.2059	0.0761
0.0016	51.0	146880	0.8529	0.2100	0.0776
0.0013	52.0	149760	0.8497	0.2109	0.0788
0.0014	53.0	152640	0.8656	0.2115	0.0777
0.0013	54.0	155520	0.8556	0.2068	0.0784
0.0012	55.0	158400	0.8539	0.2058	0.0777
0.0012	56.0	161280	0.8640	0.2006	0.0730
0.0011	57.0	164160	0.8455	0.2037	0.0770
0.0011	58.0	167040	0.8684	0.2068	0.0778
0.001	59.0	169920	0.8569	0.2057	0.0771
0.0009	60.0	172800	0.8669	0.2023	0.0741
0.001	61.0	175680	0.8780	0.2069	0.0785
0.0008	62.0	178560	0.8928	0.2068	0.0763
0.0007	63.0	181440	0.8991	0.2052	0.0750
0.0009	64.0	184320	0.9090	0.2093	0.0799
0.0007	65.0	187200	0.8953	0.2011	0.0754
0.0006	66.0	190080	0.9095	0.2059	0.0772
0.0006	67.0	192960	0.9042	0.2004	0.0766
0.0005	68.0	195840	0.9074	0.2063	0.0793

Framework versions

• Transformers 4.46.3
• Pytorch 2.1.0+cu118
• Datasets 3.1.0
• Tokenizers 0.20.3