MUSIB: musical score inpainting benchmark

EURASIP Journal on Audio, Speech, and Music Processing

Table 2 MUSIB evaluation on IrishFolk Dataset

IrishFolk Dataset (\(\approx\)300K samples)
Model	\(NLL \downarrow\)	\(pos_{F1} \uparrow\)	\(pAcc \uparrow\)	\(rAcc \uparrow\)	\(S_{div} \downarrow\)	\(H_{div} \downarrow\)	\(GS_{div} \downarrow\)
Anticipation-RNN	0.453 (*0.662)	0.930	0.657	0.860	0.017	0.060	0.007
InpaintNet	0.487 (*0.662)	0.860	0.517	0.750	0.013	0.174	0.024
SketchNet	0.539 (*0.516)	0.914	0.560	0.868	0.005	0.134	0.009
VLI	0.061	0.940	0.861	0.952	0.022	0.020	0.007

Values in boldface highlight the best performance among the four models on each metric