deepdml/whisper-tiny-af-mix-norm

Whisper Tiny af

This model is a fine-tuned version of openai/whisper-tiny on multiple datasets. It achieves the following results on the evaluation set:

• Loss: 1.2213
• Wer: 44.2578
• Cer: 17.8026

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: 64
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.04
• training_steps: 4100

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.7169	0.0244	100	1.7637	68.4393	26.7904
0.9216	0.0488	200	1.3055	52.9014	21.5255
0.6082	0.0732	300	1.1946	49.3158	19.3768
0.4534	0.0976	400	1.1545	47.5143	18.1954
0.3675	0.1220	500	1.1354	46.8387	18.5267
0.282	0.1463	600	1.1251	46.0939	19.8751
0.254	0.1707	700	1.1269	45.4876	18.8345
0.2055	0.1951	800	1.1248	48.9347	20.0803
0.1837	0.2195	900	1.1323	45.0199	19.4325
0.1606	0.2439	1000	1.1317	49.2118	21.8832
0.1337	0.2683	1100	1.1491	44.7601	18.6498
0.1149	0.2927	1200	1.1535	45.4530	19.5761
0.1072	0.3171	1300	1.1685	48.6056	20.2328
0.0998	0.3415	1400	1.1738	44.5695	18.5501
0.097	0.3659	1500	1.1702	44.4656	18.3068
0.0769	0.3902	1600	1.1601	47.1159	19.3709
0.084	0.4146	1700	1.1815	47.5663	19.6347
0.0664	0.4390	1800	1.1821	44.3097	18.7582
0.0652	0.4634	1900	1.1854	43.2184	18.4123
0.0609	0.4878	2000	1.1830	43.1145	17.4508
0.0565	0.5122	2100	1.1897	47.1505	19.0514
0.0589	0.5366	2200	1.2024	45.4010	18.6996
0.0552	0.5610	2300	1.1956	48.6402	20.3764
0.0551	0.5854	2400	1.1930	45.3837	18.6527
0.0551	0.6098	2500	1.1984	47.1159	18.6996
0.04	0.6341	2600	1.2092	47.4796	19.7725
0.0548	0.6585	2700	1.1981	42.7681	17.5915
0.0466	0.6829	2800	1.2144	48.1379	20.3588
0.0425	0.7073	2900	1.2051	46.0766	18.7670
0.0431	0.7317	3000	1.2157	44.3963	17.4596
0.0427	0.7561	3100	1.2178	48.1032	19.8517
0.0346	0.7805	3200	1.2177	47.4970	19.5644
0.0395	0.8049	3300	1.2199	47.1159	18.9312
0.039	0.8293	3400	1.2219	45.7474	19.4090
0.0359	0.8537	3500	1.2191	46.4057	18.7846
0.0461	0.8780	3600	1.2172	51.2039	21.9476
0.0299	0.9024	3700	1.2202	47.5316	19.1335
0.028	0.9268	3800	1.2216	47.1505	19.4999
0.0305	1.01	3900	1.2241	46.5443	18.7231
0.038	1.0344	4000	1.2218	44.2924	17.6619
0.0249	1.0588	4100	1.2213	44.2578	17.8026

Framework versions

• Transformers 4.42.0.dev0
• Pytorch 2.3.0+cu121
• Datasets 2.19.1
• Tokenizers 0.19.1

Citation

Please cite the model using the following BibTeX entry:

@misc{deepdml/whisper-tiny-af-mix-norm,
      title={Fine-tuned Whisper tiny ASR model for speech recognition in Afrikaans},
      author={Jimenez, David},
      howpublished={\url{https://huggingface.co/deepdml/whisper-tiny-af-mix-norm}},
      year={2026}
    }