Model description

This model was generated as part of the HF Audio course, I enjoyed it and currently this architecture achieves an amazing accuracy of 0.9 on music-genre classification task.

The Audio Spectrogram Transformer is equivalent to ViT, but applied on audio. Audio is first turned into an image (as a spectrogram), after which a Vision Transformer is applied. The model gets state-of-the-art results on several audio classification benchmarks.

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: 4
eval_batch_size: 4
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: 10
mixed_precision_training: Native AMP
global_step: 2250
training_loss: 0.23970948094350752
train_runtime: 1982.7909
train_samples_per_second: 4.534
train_steps_per_second: 1.135
total_flos: 6.094112254328832e+17
train_loss: 0.23970948094350752

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.9734	1.0	225	0.6194	0.82
0.7734	2.0	450	0.4650	0.86
0.7703	3.0	675	0.8101	0.78
0.0052	4.0	900	0.5021	0.89
0.2316	5.0	1125	0.4968	0.9
0.0001	6.0	1350	0.5484	0.87
0.5337	7.0	1575	0.4673	0.89
0.0	8.0	1800	0.4868	0.89
0.0	9.0	2025	0.4709	0.9
0.0	10.0	2250	0.4718	0.9

Framework versions

Transformers 4.35.2
Pytorch 2.1.0+cu118
Datasets 2.15.0
Tokenizers 0.15.0

beltrewilton/ast-finetuned-audioset-10-10-0.4593-finetuned-gtzan

ast-finetuned-audioset-10-10-0.4593-finetuned-gtzan