whisper-small-Fleurs_AMMI_AFRIVOICE_LRSC-ln-1hrs-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: 1.3019
• Wer: 0.3808
• Cer: 0.1461

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: 5e-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
3.6864	1.0	39	1.6821	0.9572	0.4224
1.103	2.0	78	1.0007	0.8219	0.3945
0.5188	3.0	117	0.8848	0.7838	0.3935
0.2456	4.0	156	0.8893	0.9181	0.5470
0.1259	5.0	195	0.8927	0.8016	0.4502
0.0786	6.0	234	0.9356	0.6412	0.3213
0.0595	7.0	273	0.9471	0.4937	0.1938
0.064	8.0	312	0.9619	0.5049	0.2139
0.0757	9.0	351	1.0240	0.5936	0.3168
0.0767	10.0	390	1.0239	0.4224	0.1501
0.0707	11.0	429	1.0876	0.4315	0.1580
0.0597	12.0	468	1.0901	0.4502	0.1838
0.0618	13.0	507	1.1516	0.4403	0.1795
0.0533	14.0	546	1.1576	0.4911	0.2228
0.0495	15.0	585	1.1172	0.4157	0.1594
0.0371	16.0	624	1.2017	0.4681	0.1940
0.0336	17.0	663	1.2032	0.4096	0.1584
0.0305	18.0	702	1.2111	0.4655	0.2039
0.0339	19.0	741	1.2462	0.5809	0.2864
0.0414	20.0	780	1.2423	0.4540	0.1958
0.0264	21.0	819	1.2491	0.4665	0.2037
0.0227	22.0	858	1.3357	0.4811	0.2178
0.0233	23.0	897	1.2492	0.4088	0.1621
0.0241	24.0	936	1.2684	0.4039	0.1572
0.0226	25.0	975	1.3104	0.4419	0.1942
0.0281	26.0	1014	1.3006	0.4270	0.1864
0.024	27.0	1053	1.3125	0.4359	0.1696
0.0214	28.0	1092	1.2786	0.4141	0.1667
0.0196	29.0	1131	1.2767	0.4007	0.1506
0.0224	30.0	1170	1.3100	0.3914	0.1503
0.0149	31.0	1209	1.2602	0.3949	0.1483
0.0114	32.0	1248	1.3024	0.3870	0.1457
0.0137	33.0	1287	1.3193	0.4201	0.1736
0.0154	34.0	1326	1.2994	0.4051	0.1610
0.0088	35.0	1365	1.3062	0.3989	0.1559
0.015	36.0	1404	1.2826	0.3967	0.1599
0.0055	37.0	1443	1.3145	0.3866	0.1503
0.0106	38.0	1482	1.3164	0.3952	0.1577
0.007	39.0	1521	1.3097	0.3957	0.1558
0.0035	40.0	1560	1.2840	0.3975	0.1577
0.001	41.0	1599	1.3069	0.3927	0.1560
0.0038	42.0	1638	1.2932	0.3915	0.1532
0.002	43.0	1677	1.3026	0.3911	0.1543
0.0015	44.0	1716	1.2793	0.3853	0.1485
0.0002	45.0	1755	1.2758	0.3850	0.1476
0.0005	46.0	1794	1.2775	0.3852	0.1481
0.0007	47.0	1833	1.2861	0.3861	0.1470
0.0001	48.0	1872	1.2858	0.3815	0.1425
0.0001	49.0	1911	1.2878	0.3849	0.1469
0.0001	50.0	1950	1.2892	0.3837	0.1461
0.0001	51.0	1989	1.2907	0.3835	0.1466
0.0001	52.0	2028	1.2923	0.3826	0.1462
0.0001	53.0	2067	1.2937	0.3827	0.1465
0.0001	54.0	2106	1.2951	0.3810	0.1455
0.0001	55.0	2145	1.2965	0.3810	0.1455
0.0	56.0	2184	1.2978	0.3820	0.1466
0.0	57.0	2223	1.2988	0.3817	0.1463
0.0	58.0	2262	1.3000	0.3815	0.1464
0.0	59.0	2301	1.3010	0.3811	0.1462
0.0	60.0	2340	1.3019	0.3808	0.1461

Framework versions

• Transformers 4.48.2
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0