Audio Spectrogram Transformer (AST) for speech

Adapting the Audio Spectrogram Transformer (AST) for human language classification.

Class project for Stanford CS 224S (Spoken Language Processing), spring 2024.

Notebook Pointers

Please refer to the below links to see how AST was finetuned and evaluated for various spoken language tasks.

AST for ASR

• Google FLEURS
• Librispeech

AST for Classification

• Emotion Recognition - IEMOCAP
• Emotion Classification - RAVDESS
• Music Genre Classification

Spectrogram and Attention Map Visualization

• Attention Map Visualizations