whisper-small-Fleurs_AMMI_AFRIVOICE_LRSC-ln-50hrs-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: 0.8648
• Wer: 0.2183
• Cer: 0.0778

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
1.803	1.0	1433	0.8132	0.7769	0.3908
0.7362	2.0	2866	0.5782	0.6072	0.3526
0.5456	3.0	4299	0.4854	0.7275	0.4563
0.4305	4.0	5732	0.4406	0.5160	0.3061
0.334	5.0	7165	0.4257	0.6079	0.3707
0.2498	6.0	8598	0.4294	0.4371	0.2520
0.1795	7.0	10031	0.4439	0.4619	0.2753
0.1222	8.0	11464	0.4568	0.3858	0.2024
0.0833	9.0	12897	0.4943	0.3937	0.2128
0.0605	10.0	14330	0.5112	0.3280	0.1484
0.0447	11.0	15763	0.5211	0.3067	0.1349
0.0323	12.0	17196	0.5395	0.2942	0.1250
0.0259	13.0	18629	0.5620	0.2688	0.1060
0.0202	14.0	20062	0.5688	0.2684	0.1063
0.0166	15.0	21495	0.5958	0.2737	0.1070
0.0152	16.0	22928	0.6086	0.2635	0.0931
0.0127	17.0	24361	0.6199	0.2600	0.0942
0.0109	18.0	25794	0.6309	0.2609	0.0908
0.01	19.0	27227	0.6446	0.2591	0.0930
0.0092	20.0	28660	0.6449	0.2593	0.0907
0.008	21.0	30093	0.6688	0.2481	0.0831
0.0072	22.0	31526	0.6631	0.2436	0.0836
0.0066	23.0	32959	0.6677	0.2595	0.0925
0.0065	24.0	34392	0.6984	0.2492	0.0888
0.0056	25.0	35825	0.7061	0.2434	0.0832
0.005	26.0	37258	0.7019	0.2402	0.0826
0.0055	27.0	38691	0.7125	0.2402	0.0815
0.0046	28.0	40124	0.7261	0.2385	0.0816
0.0041	29.0	41557	0.7352	0.2377	0.0831
0.0043	30.0	42990	0.7332	0.2403	0.0826
0.0037	31.0	44423	0.7439	0.2412	0.0832
0.0035	32.0	45856	0.7460	0.2369	0.0824
0.0033	33.0	47289	0.7579	0.2349	0.0819
0.0029	34.0	48722	0.7684	0.2356	0.0818
0.0029	35.0	50155	0.7658	0.2356	0.0829
0.0032	36.0	51588	0.7754	0.2292	0.0823
0.0023	37.0	53021	0.7624	0.2275	0.0805
0.0027	38.0	54454	0.7865	0.2281	0.0816
0.0024	39.0	55887	0.7885	0.2290	0.0810
0.0024	40.0	57320	0.7819	0.2296	0.0802
0.0022	41.0	58753	0.7875	0.2263	0.0792
0.0021	42.0	60186	0.7929	0.2270	0.0802
0.0024	43.0	61619	0.7969	0.2324	0.0822
0.0018	44.0	63052	0.7989	0.2338	0.0858
0.0019	45.0	64485	0.8155	0.2318	0.0849
0.0016	46.0	65918	0.8111	0.2265	0.0800
0.0016	47.0	67351	0.7933	0.2223	0.0776
0.0016	48.0	68784	0.8118	0.2232	0.0785
0.0013	49.0	70217	0.8045	0.2199	0.0774
0.0015	50.0	71650	0.8136	0.2183	0.0756
0.0012	51.0	73083	0.8240	0.2188	0.0766
0.0014	52.0	74516	0.8266	0.2198	0.0781
0.001	53.0	75949	0.8334	0.2215	0.0780
0.0013	54.0	77382	0.8225	0.2213	0.0793
0.0011	55.0	78815	0.8296	0.2190	0.0783
0.0009	56.0	80248	0.8354	0.2174	0.0772
0.0008	57.0	81681	0.8091	0.2154	0.0781
0.0009	58.0	83114	0.8459	0.2148	0.0751
0.001	59.0	84547	0.8301	0.2156	0.0755
0.0008	60.0	85980	0.8385	0.2134	0.0753
0.0008	61.0	87413	0.8474	0.2180	0.0794
0.0008	62.0	88846	0.8553	0.2170	0.0788
0.0006	63.0	90279	0.8357	0.2154	0.0756
0.0007	64.0	91712	0.8495	0.2163	0.0772
0.0006	65.0	93145	0.8508	0.2182	0.0769
0.0006	66.0	94578	0.8625	0.2226	0.0779
0.0006	67.0	96011	0.8718	0.2150	0.0763
0.0004	68.0	97444	0.8600	0.2191	0.0790
0.0005	69.0	98877	0.8622	0.2181	0.0779
0.0005	70.0	100310	0.8617	0.2205	0.0771
0.0003	71.0	101743	0.8703	0.2177	0.0790
0.0002	72.0	103176	0.8648	0.2183	0.0778

Framework versions

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