Audio XAI Attack: Adversarial Examples for Three Audio Models

Adversarial attack data for three audio models: Vggish, AST, and SpecTTTra. The dataset is hosted on Kaggle, but details on its size, creation date, and author are not provided. Its primary purpose is to explore vulnerabilities in audio machine learning models.

Use Cases

• Benchmarking model robustness based on adversarial examples for Vggish, AST, and SpecTTTra.
• Developing new adversarial attack methods using the provided audio attack data.
• Studying explainable AI (XAI) techniques for audio models under attack conditions.
• Evaluating the transferability of adversarial examples across different audio model architectures.

Strengths

• Focuses on a specific and relevant research area: adversarial attacks on audio models.
• Targets three distinct, named audio models (Vggish, AST, SpecTTTra) for comparative analysis.

Limitations

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

Provenance

Source

Kaggle