Audio Attribution on FMA Large (test) benchmark leaderboard · SOTA2 Research

Audio Attribution on FMA Large (test)

100R@1GRAD-COS

Evaluation Results

Method	Links
GRAD-COS2026.05		100	0	100	0.997	2.81	14.2	5.5	8.56	99.2	97.6	29.56	99.8	99.7
TRAK2026.05		99.3	0.4	2.2	0.991	2.78	0.1	0	5.89	0	0	7.58	0	0
CLAP2026.05		92.4	5.4	80.1	0.584	1.38	60.9	39.8	2.75	30	17.2	1.5	59.3	48.2
FACTGRASS2026.05		86.6	8.1	0	0.663	0.8	15.2	0	3.57	0	0	4.81	0	0
GRAD-COS2026.05		79.3	20	1.5	0.736	0.35	28.5	0.4	5.54	0	0	10.53	0.7	0.1
TRAK2026.05		77.2	20.8	0.2	0.641	0.45	76.1	0.4	1.25	93.9	17.5	0.73	89.1	1.6
CLEWS2026.05		60.7	14.1	40.6	0.28	0.5	51	26.3	3.04	34.9	24.8	0.71	54.4	38.7
MERT2026.05		43.7	37.5	0.2	0.358	1.09	60.4	39.5	6.55	86.8	78.2	4.84	84.5	74.7
GRAD-COS2026.05		41.3	41.9	21.6	0.374	0.36	57.6	18.6	2.15	21.7	5.7	0.93	67.1	30.5
TRAK2026.05		28.2	9.8	17.5	0.076	1.51	85.8	38.1	3.63	1.5	0	0.86	28.3	3.3
LOGRA2026.05		22.3	6.5	0	0.272	0.77	22.4	0.6	3.9	0	0	4.45	0	0
LOGRA2026.05		5.7	3.8	0.6	0.049	0.38	42.7	21.7	3.04	23.2	13.4	1.23	52.1	39.6
FACTGRASS2026.05		5.1	3	0.1	0.046	0.57	36.9	14	3.45	13.2	5.6	1.6	32.6	21.6
LOGRA2026.05		0.8	1.7	0	0.013	0.96	61.3	36.8	1.53	30.7	15.1	1.41	59.7	44.4
FACTGRASS2026.05		0.6	1.8	0	0.012	1.08	61.2	40.2	0.83	39.3	22.6	2.47	64.5	51.5