oosawy/wav2vec2-base-timit-ft

repo_name

This model is a fine-tuned version of facebook/wav2vec2-base on the timit_asr dataset. It achieves the following results on the evaluation set:

• Loss: 0.5351
• Wer: 0.2211

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: 0.0001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• 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: 1000
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.5252	1.0040	500	1.6991	0.9701
0.854	2.0080	1000	0.5187	0.4025
0.4211	3.0120	1500	0.4289	0.3326
0.2871	4.0161	2000	0.3947	0.2896
0.2266	5.0201	2500	0.4034	0.2881
0.1789	6.0241	3000	0.4833	0.2926
0.1638	7.0281	3500	0.4342	0.2776
0.15	8.0321	4000	0.4643	0.2750
0.1251	9.0361	4500	0.4449	0.2642
0.1064	10.0402	5000	0.4785	0.2578
0.0986	11.0442	5500	0.4480	0.2627
0.0883	12.0482	6000	0.4876	0.2603
0.0784	13.0522	6500	0.5100	0.2519
0.0721	14.0562	7000	0.4795	0.2536
0.0696	15.0602	7500	0.4797	0.2456
0.0598	16.0643	8000	0.5064	0.2410
0.0575	17.0683	8500	0.5075	0.2362
0.0508	18.0723	9000	0.5062	0.2420
0.048	19.0763	9500	0.5078	0.2397
0.0402	20.0803	10000	0.5511	0.2341
0.0429	21.0843	10500	0.4835	0.2330
0.0362	22.0884	11000	0.5800	0.2308
0.0333	23.0924	11500	0.5250	0.2306
0.0285	24.0964	12000	0.5242	0.2288
0.0296	25.1004	12500	0.4995	0.2238
0.0264	26.1044	13000	0.5296	0.2236
0.0245	27.1084	13500	0.5530	0.2233
0.0214	28.1124	14000	0.5376	0.2209
0.0214	29.1165	14500	0.5351	0.2211

Framework versions

• Transformers 4.56.2
• Pytorch 2.8.0+cu126
• Datasets 2.21.0
• Tokenizers 0.22.1