Practically Efficient Blind Speech Separation Using Frequency Band Selection Based on Magnitude Squared Coherence and a Small Dodecahedral Microphone Array

Table 3

Estimated computational complexity.

Method	Complexity; (conventional)	Complexity; (proposed)	Ratio; (reduction (%))	Complexity; (TRINICON)	Ratio; (reduction (%))
STFT (forward and inverse)	7.68𝐸8	7.68𝐸8	1.0 (0 [%])	3.69𝐸9	4.8
Covariance matrix	1.80𝐸9	7.20𝐸8	0.4 (60 [%])	4.61𝐸10	25.6
Eigenvalue decomposition	7.49𝐸8	3.00𝐸8	0.4 (60 [%])	7.67𝐸9	10.3
Subspace method	7.50𝐸8	3.00𝐸8	0.4 (60 [%])	4.01𝐸10	53.5
Separation matrix	2.50𝐸9	1.00𝐸9	0.4 (60 [%])	3.13𝐸11	156.6
Projection method	7.49𝐸8	3.00𝐸8	0.4 (60 [%])	7.67𝐸9	10.3
Hierarchical clustering	5.18𝐸11	8.29𝐸10	0.16 (84 [%])	—	—
Total	5.26𝐸11	8.63𝐸10	0.16 (84 [%])	4.19𝐸11	0.80 (20 [%])