deepdml/whisper-medium-af-mix-norm

Whisper Medium af

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

• Loss: 0.6508
• Wer: 21.7911
• Cer: 7.5309

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: 16
• eval_batch_size: 16
• 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
0.6031	0.0244	100	0.7367	30.9198	11.1248
0.3567	0.0488	200	0.6171	28.3908	11.3388
0.3101	0.0732	300	0.5763	24.5626	8.7914
0.1999	0.0976	400	0.5673	24.2855	8.3194
0.1453	0.1220	500	0.5739	23.0210	8.4191
0.1184	0.1463	600	0.5742	24.6666	8.2960
0.116	0.1707	700	0.5686	25.7059	9.6532
0.0941	0.1951	800	0.5905	23.8524	7.9618
0.1025	0.2195	900	0.6025	24.8398	9.4216
0.0889	0.2439	1000	0.5666	23.0383	7.9970
0.0492	0.2683	1100	0.5936	22.9863	8.0761
0.0554	0.2927	1200	0.6092	22.8477	8.0732
0.0448	0.3171	1300	0.6118	26.4854	10.9694
0.0358	0.3415	1400	0.6163	25.9484	10.0314
0.0454	0.3659	1500	0.6139	23.0729	7.8768
0.0286	0.3902	1600	0.6177	22.5533	7.9354
0.0395	0.4146	1700	0.6100	25.7752	10.1134
0.058	0.4390	1800	0.6200	23.1942	9.1607
0.0269	0.4634	1900	0.6309	23.1076	8.2168
0.0229	0.4878	2000	0.6334	23.3674	8.7474
0.0394	0.5122	2100	0.6262	22.8131	7.7976
0.0186	0.5366	2200	0.6295	25.0476	9.7060
0.0267	0.5610	2300	0.6427	22.9517	8.4718
0.0161	0.5854	2400	0.6334	23.0902	8.3986
0.023	0.6098	2500	0.6368	21.8604	7.6071
0.028	0.6341	2600	0.6319	22.0336	7.6540
0.0237	0.6585	2700	0.6296	22.9517	7.9002
0.0152	0.6829	2800	0.6601	22.3627	7.6129
0.0143	0.7073	2900	0.6544	21.8431	7.4605
0.0151	0.7317	3000	0.6541	22.5013	7.5191
0.0157	0.7561	3100	0.6583	22.6745	7.6862
0.0096	0.7805	3200	0.6594	22.4840	7.6569
0.0124	0.8049	3300	0.6476	21.8431	7.6716
0.02	0.8293	3400	0.6406	21.9816	7.7390
0.0204	0.8537	3500	0.6399	21.4966	7.6041
0.018	0.8780	3600	0.6446	21.8777	7.6979
0.0125	0.9024	3700	0.6585	21.8777	7.4927
0.0095	0.9268	3800	0.6601	21.9124	7.4752
0.0076	0.9512	3900	0.6571	22.0509	7.5279
0.0176	0.9756	4000	0.6521	21.7565	7.5191
0.0086	1.0	4100	0.6508	21.7911	7.5309

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