afrivoices-whisper-turbo-50h

This model is a fine-tuned version of openai/whisper-large-v3-turbo on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.6472
• Wer: 0.2822

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 200
• num_epochs: 3
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.3245	0.0926	500	1.2512	1.1332
2.0196	0.1852	1000	1.1085	0.9513
1.9094	0.2778	1500	0.9980	0.5464
1.7282	0.3704	2000	0.9755	0.5172
1.5461	0.4630	2500	0.9266	0.5150
1.5012	0.5556	3000	0.9051	0.4016
1.4850	0.6482	3500	0.8571	0.3469
1.5167	0.7408	4000	0.8240	0.3211
1.4645	0.8334	4500	0.8260	0.3582
1.4691	0.9260	5000	0.7812	0.4899
1.2002	1.0185	5500	0.7821	0.5344
1.1558	1.1111	6000	0.7543	0.4379
1.1442	1.2037	6500	0.7418	0.5382
1.2177	1.2963	7000	0.7233	0.5535
1.1813	1.3889	7500	0.7229	0.3162
1.1705	1.4815	8000	0.7128	0.2949
1.1547	1.5741	8500	0.6989	0.3372
1.1229	1.6667	9000	0.6993	0.3215
1.1229	1.7593	9500	0.6784	0.3013
1.0661	1.8519	10000	0.6844	0.2725
1.0986	1.9445	10500	0.6725	0.3608
0.9513	2.0370	11000	0.6840	0.2766
0.8773	2.1296	11500	0.6763	0.2972
0.9462	2.2222	12000	0.6605	0.2766
0.9006	2.3148	12500	0.6592	0.2668
0.9189	2.4074	13000	0.6666	0.2642
0.9381	2.5000	13500	0.6559	0.2818
0.8477	2.5926	14000	0.6580	0.3050
0.8142	2.6852	14500	0.6514	0.3088
0.8875	2.7778	15000	0.6503	0.2762
0.8470	2.8705	15500	0.6446	0.2736
0.7688	2.9631	16000	0.6476	0.2814
0.7720	3.0	16200	0.6472	0.2822

Framework versions

• Transformers 5.12.1
• Pytorch 2.11.0+cu128
• Datasets 5.0.0
• Tokenizers 0.22.2