Here we provide the source code for the paper "Graph Attention for Automated Audio Captioning" accepted by IEEE Signal Processing Letter.
The model structure of the proposed GraphAC is provided in the file path model, and the dependent module parts (i.e., PANNs, GAT and the SpecAugment operation) are provided in the file path modules.
The example of predicted captions is presented at captions_of_examples.md, accompanying Figure 2 of the paper "Graph Attention for Automated Audio Captioning". It demonstrates that the proposed method can accurately capture and caption the long-time dependent information.
This project is released under the CC BY-NC-ND 4.0 license.