A music-aware transformer framework combining multi-scale spectrogram learning and adaptive attention optimization reported ...
A computational study of GTZAN and ASAP audio datasets reports that the CCTAF-MSR CNN–Transformer model reached 86.40% ...