teckedd/whisper_small-waxal_akan-asr

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: 0.4664
• Wer: 33.9911

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: 100
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.898	0.1	100	0.8296	53.0885
0.5673	0.2	200	0.6036	41.1596
0.5163	0.3	300	0.5389	38.9718
0.4647	1.084	400	0.5052	36.3355
0.4073	1.184	500	0.4919	35.4310
0.3885	1.284	600	0.4835	35.1604
0.3824	2.068	700	0.4720	34.1206
0.3513	2.168	800	0.4709	34.6946
0.3521	2.268	900	0.4675	34.4240
0.335	3.052	1000	0.4664	33.9911

Framework versions

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