whisper-small-Fleurs_AMMI_AFRIVOICE_LRSC-ln-5hrs-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: 0.9246
• Wer: 0.2946
• Cer: 0.1160

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
3.2458	1.0	197	1.3723	1.0300	0.5178
1.1335	2.0	394	0.9104	0.8444	0.4119
0.7565	3.0	591	0.7383	0.7341	0.3627
0.5358	4.0	788	0.6557	0.7250	0.3796
0.3661	5.0	985	0.6261	0.8424	0.4895
0.2372	6.0	1182	0.6162	0.7220	0.4038
0.1402	7.0	1379	0.6452	0.8728	0.5281
0.0823	8.0	1576	0.6453	0.9265	0.5762
0.0506	9.0	1773	0.6683	0.7950	0.4716
0.038	10.0	1970	0.6756	0.8420	0.5188
0.0288	11.0	2167	0.7030	0.6811	0.3833
0.0218	12.0	2364	0.7101	0.4698	0.2265
0.0156	13.0	2561	0.7235	0.6326	0.3551
0.0119	14.0	2758	0.7426	0.4402	0.2030
0.0104	15.0	2955	0.7523	0.5167	0.2593
0.0073	16.0	3152	0.7586	0.4773	0.2402
0.0062	17.0	3349	0.7481	0.4544	0.2327
0.004	18.0	3546	0.7596	0.4101	0.1972
0.0022	19.0	3743	0.7582	0.3821	0.1676
0.0023	20.0	3940	0.7786	0.4205	0.2031
0.0021	21.0	4137	0.7690	0.4023	0.1838
0.0021	22.0	4334	0.7817	0.4109	0.1830
0.0016	23.0	4531	0.7935	0.4091	0.1793
0.0021	24.0	4728	0.7997	0.4872	0.2415
0.0045	25.0	4925	0.7827	0.3801	0.1615
0.0065	26.0	5122	0.8009	0.3772	0.1667
0.0056	27.0	5319	0.7998	0.3757	0.1680
0.0053	28.0	5516	0.8201	0.3383	0.1376
0.0045	29.0	5713	0.8074	0.4136	0.1972
0.0033	30.0	5910	0.8100	0.3300	0.1318
0.0012	31.0	6107	0.8238	0.3600	0.1674
0.0013	32.0	6304	0.8251	0.3342	0.1314
0.0008	33.0	6501	0.8447	0.3166	0.1208
0.0015	34.0	6698	0.8350	0.3079	0.1159
0.001	35.0	6895	0.8499	0.3329	0.1364
0.0021	36.0	7092	0.8348	0.3158	0.1225
0.0012	37.0	7289	0.8511	0.2980	0.1062
0.0014	38.0	7486	0.8434	0.3392	0.1386
0.0018	39.0	7683	0.8632	0.3224	0.1284
0.0021	40.0	7880	0.8481	0.3721	0.1738
0.0021	41.0	8077	0.8448	0.3692	0.1645
0.0033	42.0	8274	0.8678	0.3424	0.1474
0.0011	43.0	8471	0.8617	0.3218	0.1269
0.0012	44.0	8668	0.8570	0.2992	0.1180
0.0007	45.0	8865	0.8609	0.2970	0.1115
0.0008	46.0	9062	0.8931	0.3132	0.1185
0.001	47.0	9259	0.8838	0.3127	0.1291
0.0005	48.0	9456	0.8732	0.2906	0.1145
0.0026	49.0	9653	0.8671	0.2949	0.1120
0.0014	50.0	9850	0.8750	0.3150	0.1308
0.0009	51.0	10047	0.8786	0.3084	0.1292
0.0007	52.0	10244	0.8850	0.3007	0.1132
0.0005	53.0	10441	0.8919	0.2918	0.1058
0.0003	54.0	10638	0.8893	0.2980	0.1156
0.0002	55.0	10835	0.9022	0.2953	0.1155
0.0	56.0	11032	0.9002	0.2849	0.1020
0.0005	57.0	11229	0.9138	0.2853	0.1032
0.0007	58.0	11426	0.8995	0.2957	0.1158
0.0004	59.0	11623	0.8997	0.2854	0.1061
0.0013	60.0	11820	0.9066	0.2837	0.0993
0.002	61.0	12017	0.9002	0.3039	0.1150
0.001	62.0	12214	0.9220	0.2925	0.1088
0.0014	63.0	12411	0.9129	0.3056	0.1212
0.0009	64.0	12608	0.9137	0.2844	0.1051
0.0008	65.0	12805	0.9128	0.3075	0.1307
0.0005	66.0	13002	0.9248	0.2870	0.1040
0.0004	67.0	13199	0.9281	0.2855	0.1011
0.0007	68.0	13396	0.9097	0.2883	0.1079
0.0005	69.0	13593	0.9092	0.2981	0.1158
0.0004	70.0	13790	0.9170	0.2977	0.1133
0.0002	71.0	13987	0.9252	0.3131	0.1272
0.001	72.0	14184	0.9246	0.2946	0.1160

Framework versions

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