deepdml/whisper-base-af-mix-norm

Whisper Base af

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

• Loss: 0.9827
• Wer: 36.8093
• Cer: 13.8188

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: 2100

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.0677	0.0476	100	1.1941	49.7488	18.2687
0.5325	0.0952	200	0.9858	44.6735	17.1548
0.3382	0.1429	300	0.9492	41.2264	15.9675
0.2433	0.1905	400	0.9351	38.5761	14.4520
0.197	0.2381	500	0.9300	37.5368	13.9419
0.1455	0.2857	600	0.9291	43.1318	17.5153
0.1339	0.3333	700	0.9403	43.2011	18.0371
0.1041	0.3810	800	0.9469	41.2437	16.4160
0.0968	0.4286	900	0.9484	38.1431	14.5575
0.0841	0.4762	1000	0.9601	38.3336	15.0676
0.0725	0.5238	1100	0.9554	39.7367	16.5157
0.0621	0.5714	1200	0.9637	37.6754	14.4461
0.0598	0.6190	1300	0.9665	36.9998	14.6836
0.056	0.6667	1400	0.9735	41.4689	16.0965
0.0554	0.7143	1500	0.9761	37.3116	13.8422
0.0421	0.7619	1600	0.9784	41.6075	15.9558
0.0504	0.8095	1700	0.9813	37.0171	13.8657
0.0428	0.8571	1800	0.9841	36.9998	14.2145
0.0426	0.9048	1900	0.9847	40.6375	15.7242
0.0397	0.9524	2000	0.9820	37.0518	13.9272
0.0383	1.0	2100	0.9827	36.8093	13.8188

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