Abstract
Increasing depth of transformer models improves recognition, but it comes at a substantial cost. Each additional layer requires more parameters, which makes the process computationally inefficient. We ask whether additional processing can focus on integrating features already computed. Looped Audio Spectrogram Transformer (LAST) first processes all tokens, then reuses the same blocks to refine only the class token over fixed audio features, thereby making later passes inexpensive. On AudioSet, ten-pass LAST achieves 0.345 mean average precision, exceeding a twelve-layer sequential transformer by 2.1% relative with 49.4% fewer parameters, 42% fewer multiply-accumulate operations, and 9.8% higher measured throughput. Across separately trained models, increasing the pass count from two to ten improves accuracy while adding only 1.2% computation. Further evaluations show improved robustness to temporal masking and various other auditory augmentations, with better generalization on classification tasks with music, environmental, and event sounds.
Keywords
Subject
Publication details
- Journal
- Not available
- Open access
- Green open access
Cite this article
APA 7
Al-Tahan, H., O'Brien, S., Razdaibiedina, A., & Murty, N. A. R. (2026). LAST: Looped Audio Spectrogram Transformer. https://omanscience.com/en/articles/last-looped-audio-spectrogram-transformer
MLA 9
Al-Tahan, Haider, et al. "LAST: Looped Audio Spectrogram Transformer." https://omanscience.com/en/articles/last-looped-audio-spectrogram-transformer.
Chicago (author–date)
Al-Tahan, Haider, Sean O'Brien, Anastasia Razdaibiedina, and N. Apurva Ratan Murty. 2026. "LAST: Looped Audio Spectrogram Transformer." https://omanscience.com/en/articles/last-looped-audio-spectrogram-transformer.
Harvard
Al-Tahan, H., O'Brien, S., Razdaibiedina, A. and Murty, N. A. R. (2026) 'LAST: Looped Audio Spectrogram Transformer', Available at: https://omanscience.com/en/articles/last-looped-audio-spectrogram-transformer.
Vancouver
Al-Tahan H, O'Brien S, Razdaibiedina A, Murty NAR. LAST: Looped Audio Spectrogram Transformer. https://omanscience.com/en/articles/last-looped-audio-spectrogram-transformer
IEEE
H. Al-Tahan, S. O'Brien, A. Razdaibiedina, and N. A. R. Murty, "LAST: Looped Audio Spectrogram Transformer," https://omanscience.com/en/articles/last-looped-audio-spectrogram-transformer.