ai_music_detection_large_60s

This model was trained from mit/ast-finetuned-audioset-10-10-0.4593 on the SleepyJesse/ai_music_large dataset.

Please see the code in the Jupyter Notebook in files.

Model description

The model was trained with max_length = 6000, which is 60 seconds.

Intended uses & limitations

This model is used to classify a given music piece is AI-generated or human-composed.

Training and evaluation data

The SleepyJesse/ai_music_large dataset was used, with 80% train/test split, and 0.8 probability for audio data augmentation.

Training procedure

See ai_music_detection_new_large_60.ipynb and training metrics.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 2
• eval_batch_size: 16
• seed: 42
• 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: linear
• num_epochs: 10

Framework versions

• Transformers 4.46.3
• Pytorch 2.5.1
• Datasets 3.1.0
• Tokenizers 0.20.3