deepdml/whisper-small-af-mix-norm

Whisper Small af

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

• Loss: 0.7741
• Wer: 26.9531
• Cer: 9.2135

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: 32
• eval_batch_size: 32
• 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
0.8702	0.0244	100	0.9743	39.9099	14.1676
0.4455	0.0488	200	0.7622	31.5954	11.4297
0.2706	0.0732	300	0.7169	31.7513	11.7668
0.1712	0.0976	400	0.7184	29.6380	10.0871
0.1377	0.1220	500	0.6963	27.8365	10.0284
0.0942	0.1463	600	0.7097	28.2002	9.9786
0.0727	0.1707	700	0.7194	27.8018	9.7089
0.0586	0.1951	800	0.7261	27.8192	9.6620
0.0756	0.2195	900	0.7278	27.3515	9.6210
0.0508	0.2439	1000	0.7276	27.6286	9.4773
0.0377	0.2683	1100	0.7395	27.8711	9.6943
0.035	0.2927	1200	0.7406	28.0963	9.4392
0.0433	0.3171	1300	0.7443	27.8192	9.6122
0.0339	0.3415	1400	0.7526	28.4254	9.7998
0.033	0.3659	1500	0.7475	30.3828	10.8756
0.0203	0.3902	1600	0.7610	27.1090	9.0640
0.0371	0.4146	1700	0.7445	27.0397	9.0757
0.0271	0.4390	1800	0.7529	26.9357	9.3395
0.0213	0.4634	1900	0.7776	27.4034	9.2194
0.024	0.4878	2000	0.7596	27.3341	9.4070
0.0273	0.5122	2100	0.7700	27.6459	9.5389
0.0221	0.5366	2200	0.7744	27.2822	9.3014
0.0115	0.5610	2300	0.7754	29.6899	10.8786
0.0156	0.5854	2400	0.7695	27.4381	9.5008
0.0183	0.6098	2500	0.7658	27.2822	9.3249
0.0156	0.6341	2600	0.7801	27.5074	9.4509
0.0093	0.6585	2700	0.7717	26.8664	9.2018
0.0127	0.6829	2800	0.7876	28.7892	10.5854
0.016	0.7073	2900	0.7695	26.9357	9.3044
0.0161	0.7317	3000	0.7827	27.1956	9.1080
0.0094	0.7561	3100	0.7738	27.1609	9.3249
0.0071	0.7805	3200	0.7895	27.0223	9.1959
0.0106	0.8049	3300	0.7844	29.1010	10.6294
0.014	0.8293	3400	0.7698	29.3781	10.6558
0.0108	0.8537	3500	0.7786	27.1090	9.2164
0.0088	0.8780	3600	0.7778	27.1436	9.2223
0.0117	0.9024	3700	0.7758	27.1956	9.3659
0.0099	0.9268	3800	0.7756	26.9357	9.2340
0.0104	0.9512	3900	0.7769	27.0223	9.1959
0.0118	0.9756	4000	0.7738	26.6759	9.1637
0.0104	1.0	4100	0.7741	26.9531	9.2135

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