wav2vec2-base-960h-fsc

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

• Loss: 0.0218
• Accuracy: 0.9947

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: 0.0005
• train_batch_size: 48
• eval_batch_size: 48
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 192
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 25

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	0.9959	120	0.3651	0.9380
No log	2.0	241	0.2352	0.9404
No log	2.9959	361	0.4245	0.8684
No log	4.0	482	0.0721	0.9837
No log	4.9959	602	0.0961	0.9768
No log	6.0	723	0.0632	0.9860
No log	6.9959	843	0.0498	0.9905
No log	8.0	964	0.0715	0.9834
0.4012	8.9959	1084	0.0907	0.9829
0.4012	10.0	1205	0.0644	0.9860
0.4012	10.9959	1325	0.0322	0.9921
0.4012	12.0	1446	0.0524	0.9881
0.4012	12.9959	1566	0.0450	0.9910
0.4012	14.0	1687	0.0227	0.9942
0.4012	14.9959	1807	0.0437	0.9908
0.4012	16.0	1928	0.0381	0.9924
0.1096	16.9959	2048	0.0218	0.9947
0.1096	18.0	2169	0.0300	0.9934
0.1096	18.9959	2289	0.0356	0.9931
0.1096	20.0	2410	0.0380	0.9937
0.1096	20.9959	2530	0.0417	0.9934
0.1096	22.0	2651	0.0268	0.9947

Framework versions

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