speecht5-tunis_finalll

This model is a fine-tuned version of microsoft/speecht5_asr on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2003
• Wer Ortho: 62.9526
• Wer: 59.7855

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: 5e-06
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• 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: cosine
• lr_scheduler_warmup_steps: 50
• training_steps: 1500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Wer Ortho
1.7381	0.3731	100	1.4437	213.9108	371.3889
0.8437	0.7463	200	0.5686	81.6273	80.5556
0.4461	1.1194	300	0.3668	77.4278	76.1111
0.3753	1.4925	400	0.2760	74.0157	72.7778
0.3416	1.8657	500	0.2392	84.7769	80.8333
0.2656	2.2388	600	0.2138	67.9790	67.7778
0.2706	2.6119	700	0.2085	77.1654	74.7222
0.2509	2.9851	800	0.1995	62.2047	63.0556
0.2314	3.3582	900	0.1949	61.6798	62.5
0.2806	3.7313	1000	0.1951	62.4672	63.3333
0.2254	4.1045	1100	0.1912	68.6111	69.2913
0.2674	4.4776	1200	0.1863	68.6111	69.8163
0.301	4.8507	1300	0.1862	67.5	67.9790
0.2354	5.2239	1400	0.1850	61.1111	59.8425
0.2349	5.5970	1500	0.1851	67.2222	67.7165

Framework versions

• Transformers 4.52.4
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.2