datdo/wav2vec2-base-timit-demo-google-colab-clone

wav2vec2-base-timit-demo-google-colab

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: 0.5720
• Wer: 0.3380

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: Adam with betas=(0.9,0.999) and epsilon=1e-08
• 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
3.466	1.0040	500	1.4947	0.9858
0.8266	2.0080	1000	0.5298	0.5179
0.438	3.0120	1500	0.4565	0.4564
0.2918	4.0161	2000	0.4528	0.4382
0.2282	5.0201	2500	0.4541	0.4095
0.184	6.0241	3000	0.5109	0.4053
0.1513	7.0281	3500	0.5116	0.3923
0.1378	8.0321	4000	0.5137	0.3876
0.1194	9.0361	4500	0.5208	0.3961
0.1072	10.0402	5000	0.5417	0.3845
0.0982	11.0442	5500	0.5653	0.3847
0.0868	12.0482	6000	0.4593	0.3722
0.0774	13.0522	6500	0.4822	0.3723
0.0723	14.0562	7000	0.5303	0.3702
0.0635	15.0602	7500	0.4888	0.3742
0.0597	16.0643	8000	0.5254	0.3638
0.0571	17.0683	8500	0.5107	0.3632
0.0491	18.0723	9000	0.5649	0.3575
0.0511	19.0763	9500	0.5430	0.3627
0.0425	20.0803	10000	0.5726	0.3633
0.0386	21.0843	10500	0.5977	0.3657
0.0388	22.0884	11000	0.5430	0.3570
0.0338	23.0924	11500	0.5612	0.3535
0.0301	24.0964	12000	0.5841	0.3514
0.0272	25.1004	12500	0.5682	0.3457
0.0255	26.1044	13000	0.5657	0.3494
0.0234	27.1084	13500	0.5611	0.3450
0.0248	28.1124	14000	0.5721	0.3399
0.0203	29.1165	14500	0.5720	0.3380

Framework versions

• Transformers 4.41.2
• Pytorch 2.1.2
• Datasets 2.19.2
• Tokenizers 0.19.1