IOAI 2026 Home Task 1 Audio Dataset: AST Checkpoint and Sound Classes

IOAI 2026 Home Task 1 Audio Dataset provides audio data and an Audio Spectrogram Transformer checkpoint for 29 sound classes. The dataset includes 16 retained and 13 new sound classes. It was created for a Kaggle competition task, though the specific author and organization are unknown.

Use Cases

• Fine-tuning Audio Spectrogram Transformer models based on the provided checkpoint.
• Training audio classifiers for the 29 specified sound classes.
• Benchmarking model performance on a mix of retained and new audio categories.
• Developing and evaluating audio data augmentation techniques.

Strengths

• Includes a pre-trained Audio Spectrogram Transformer checkpoint, providing a starting point for model development.
• Covers a defined set of 29 sound classes, offering a clear classification target.
• Distinguishes between 16 retained and 13 new classes, which may be useful for transfer learning studies.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count, file formats, and license are unknown, which may limit suitability assessment.

Provenance

Source

Kaggle