wav2vec2-base-960h-speech-commands

This model is a fine-tuned version of facebook/wav2vec2-base-960h on the speech_commands dataset. It achieves the following results on the evaluation set:

• Loss: 1.1612
• Accuracy: 0.8067

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-05
• train_batch_size: 48
• eval_batch_size: 48
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 20
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
1.745	1.0	824	1.9237	0.7648
0.5664	2.0	1648	1.1424	0.7878
0.4337	3.0	2472	1.1234	0.8013
0.3346	4.0	3296	1.1040	0.8035
0.2683	5.0	4120	1.3128	0.7905
0.3498	6.0	4944	1.2172	0.7972
0.2556	7.0	5768	1.1906	0.7986
0.226	8.0	6592	1.1081	0.8044
0.2317	9.0	7416	1.1068	0.8049
0.1144	10.0	8240	1.1612	0.8067
0.2143	11.0	9064	1.1577	0.8031
0.1668	12.0	9888	1.1343	0.8058
0.2504	13.0	10712	1.0583	0.8067
0.218	14.0	11536	1.0677	0.8026
0.1025	15.0	12360	1.0690	0.8053

Framework versions

• Transformers 4.43.3
• Pytorch 2.2.2+cu121
• Datasets 2.18.0
• Tokenizers 0.19.1