Table 4.

The influence of reduction ratio r on separation performance after the network layer is deepened.

Singing voice				Accompaniment
Reduction ratio r	GNSDR	GSIR	GSAR	GNSDR	GSIR	GSAR
Original	7.91	12.93	10.66	7.33	9.95	11.91
4	7.95	13.17	10.58	7.41	10.13	11.85
8	7.97	13.50	10.45	7.24	9.77	12.00
16	8.07	13.64	10.49	7.34	9.90	12.07
32	7.96	13.33	10.54	7.32	9.94	11.95