melvinrajendran/whisper-tiny-coraal

Whisper Tiny - CORAAL

This model is a fine-tuned version of openai/whisper-tiny on the AAVE Speech Recognition CORAAL dataset. It achieves the following results on the evaluation set:

• Loss: 0.7214
• Wer: 29.4196

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: 11
• 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: 50
• training_steps: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.5330	0.6579	25	1.2165	32.8886
1.0415	1.3158	50	0.8379	29.4863
0.7776	1.9737	75	0.7499	30.3536
0.6051	2.6316	100	0.7323	28.4857
0.4924	3.2895	125	0.7279	36.7578
0.4527	3.9474	150	0.7222	36.2909
0.3521	4.6053	175	0.7257	30.9540
0.3158	5.2632	200	0.7214	29.4196

Framework versions

• Transformers 5.20.1
• Pytorch 2.11.0+cu130
• Datasets 5.1.0
• Tokenizers 0.23.2