EURASIP Journal on Audio, Speech, and Music Processing

Table 2 Audio scene classification performance under different positive constraints

From: Discriminative frequency filter banks learning with neural networks

Initialization	Constraint	F-B		T-I
		Accuracy	MCC	Accuracy	MCC
N(0, 0.1)	Exponent	77.21	75.85	71.20	69.48
	Sigmoid	77.87	76.60	70.75	68.97
	ReLU	77.37	76.05	76.92	75.59
	Square	77.89	76.63	75.96	74.62
N(− 3.0, 2.0)	Exponent	77.97	76.70	73.52	72.04
	Sigmoid	77.16	75.82	71.44	69.81

F-B represents the parameters with fixed frequency center and bandwidth; T-I represents the totally independent parameters; N means Gaussian distribution

Back to article page