Skip to main content
EURASIP Journal on Audio, Speech, and Music Processing

Table 6 Ablation study of ensemble with data augmentation. Spect, Modgd, and Tempo refer to Mel-spectrogram-Swin-T, Modgdgram-Swin-T, and Tempogram-Swin-T, respectively. + denotes soft voting

From: Transformer-based ensemble method for multiple predominant instruments recognition in polyphonic music

Sl.No

Ensemble

F1 Micro

F1 Macro

1

Spect + Modgd.

0.59

0.60

2

Spect + Tempo.

0.64

0.60

3

Modgd + Tempo.

0.57

0.55

4

Spect + Modgd + Tempo.

0.66

0.62

Back to article page