teckedd/whisper_small-waxal_akan-asr-v1

Whisper Small SerendepifyLabs Twi ASR

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

• Loss: 0.4755
• Wer: 34.2849

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: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH 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
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.6845	0.1	200	0.6603	44.2810
0.4908	1.042	400	0.5130	37.1241
0.4007	1.142	600	0.4811	34.9633
0.3474	2.084	800	0.4633	34.6444
0.3016	3.026	1000	0.4581	34.0858
0.2758	3.126	1200	0.4647	34.7391
0.242	4.068	1400	0.4663	34.1438
0.2211	5.01	1600	0.4726	35.0309
0.1916	5.11	1800	0.4766	34.0607
0.1989	6.052	2000	0.4755	34.2849

Framework versions

• Transformers 4.52.0
• Pytorch 2.8.0+cu128
• Datasets 3.6.0
• Tokenizers 0.21.4