Reproduce “SSAST: Self-Supervised Audio Spectrogram Transformer” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yuan Gong, Cheng-I Jeff Lai, Yu-An Chung, James Glass

SSAST pretrains an audio spectrogram transformer on unlabeled audio by masking time-frequency patches and training it to identify and reconstruct them. Fine-tuned with labeled data, it improves over AST-Scratch across six audio and speech benchmarks. On AudioSet-20K, SSAST 400 reaches 0.310 mAP versus 0.148 for scratch, but trails the supervised AST-IM+KD baseline at 0.347.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.