whisper-small-Fleurs_AMMI_AFRIVOICE_LRSC-ln-10hrs-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: 0.9113
• Wer: 0.2693
• Cer: 0.1004

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
2.662	1.0	384	1.1130	0.9642	0.4862
0.9465	2.0	768	0.7685	0.7183	0.3554
0.6467	3.0	1152	0.6290	0.7266	0.3898
0.4612	4.0	1536	0.5702	0.6397	0.3367
0.3264	5.0	1920	0.5509	0.6423	0.3565
0.2166	6.0	2304	0.5555	0.9089	0.5573
0.136	7.0	2688	0.5589	0.9389	0.5901
0.0854	8.0	3072	0.5840	0.9759	0.5946
0.0535	9.0	3456	0.6099	0.7188	0.4253
0.0401	10.0	3840	0.6250	0.8047	0.4718
0.0305	11.0	4224	0.6397	0.5552	0.3004
0.0234	12.0	4608	0.6584	0.4350	0.2223
0.0179	13.0	4992	0.6768	0.4224	0.1984
0.0141	14.0	5376	0.6962	0.4579	0.2346
0.0109	15.0	5760	0.7025	0.4128	0.1926
0.0096	16.0	6144	0.6962	0.3778	0.1732
0.0069	17.0	6528	0.7180	0.3991	0.1853
0.0061	18.0	6912	0.7368	0.3223	0.1319
0.0045	19.0	7296	0.7343	0.3606	0.1692
0.0054	20.0	7680	0.7521	0.2899	0.1133
0.0039	21.0	8064	0.7545	0.3050	0.1204
0.0052	22.0	8448	0.7758	0.2871	0.1105
0.0041	23.0	8832	0.7674	0.2952	0.1189
0.0045	24.0	9216	0.7896	0.2945	0.1143
0.0033	25.0	9600	0.7860	0.2891	0.1149
0.0026	26.0	9984	0.8004	0.2691	0.0999
0.0029	27.0	10368	0.8039	0.2696	0.0999
0.0036	28.0	10752	0.8287	0.2634	0.0939
0.0027	29.0	11136	0.7887	0.2707	0.1002
0.0032	30.0	11520	0.8300	0.2734	0.1032
0.0029	31.0	11904	0.8122	0.2628	0.0954
0.0032	32.0	12288	0.8459	0.2664	0.0962
0.0031	33.0	12672	0.8250	0.2741	0.1075
0.0023	34.0	13056	0.8579	0.2649	0.0966
0.0023	35.0	13440	0.8535	0.2606	0.0961
0.0018	36.0	13824	0.8601	0.2571	0.0930
0.0028	37.0	14208	0.8426	0.2603	0.0969
0.0021	38.0	14592	0.8617	0.2591	0.0967
0.0016	39.0	14976	0.8570	0.2572	0.0930
0.0022	40.0	15360	0.8496	0.2581	0.0926
0.0015	41.0	15744	0.8533	0.2578	0.0958
0.0014	42.0	16128	0.8752	0.2520	0.0891
0.0014	43.0	16512	0.8737	0.2552	0.0918
0.0013	44.0	16896	0.8928	0.2616	0.0974
0.0026	45.0	17280	0.9111	0.2613	0.0957
0.0029	46.0	17664	0.8834	0.2672	0.0999
0.0018	47.0	18048	0.8904	0.2555	0.0916
0.0014	48.0	18432	0.9028	0.2541	0.0893
0.0013	49.0	18816	0.8990	0.2558	0.0904
0.0014	50.0	19200	0.9087	0.2564	0.0927
0.0014	51.0	19584	0.9115	0.2610	0.0934
0.0007	52.0	19968	0.9291	0.2592	0.0938
0.0011	53.0	20352	0.9081	0.2603	0.0970
0.0014	54.0	20736	0.9113	0.2693	0.1004

Framework versions

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