Deep Recurrent Neural Network-Based Autoencoders for Acoustic Novelty Detection

Table 1

Acoustic novel events in the test set. Shown are the number of different events per database, the average duration, and the total duration in seconds per event type. The last column indicates the total number of events and total duration across the databases. The last line indicates the total duration in seconds of the test set including normal and novel events per database.

Events	A3Novelty		PASCAL CHiME		PROMETHEUS								Total
					ATM		Corridor		Outdoor		Smart-room
	#	Time (avg.)	#	Time (avg.)	#	Time (avg.)	#	Time (avg.)	#	Time (avg.)	#	Time (avg.)	#	time
Alarm	—	—	76	435.8 (6.0)	—	—	6	84.0 (14.0)	—	—	3	9.0 (3.0)	85	528.8
Anger	—	—	—	—	—	—	—	—	6	293.0 (48.8)	—	—	6	293.0
Fall	3	4.2 (2.1)	48	89.5 (1.8)	—	—	3	3.0 (1.0)	—	—	2	2.0 (1.0)	55	98.7
Fracture	1	2.2	32	70.4 (2.2)	—	—	—	—	—	—	—	—	33	72.6
Pain	—	—	—	—	—	—	2	8.0 (4.0)	—	—	5	67.0 (13.4)	7	75.0
Scream	6	10.4 (1.7)	111	214.6 (1.9)	5	30.0 (6.0)	25	228.0 (9.1)	4	48.0 (12.0)	10	234.0 (23.4)	159	762.2
Siren	3	20.4 (6.8)	—	—	—	—	—	—	—	—	—	—	3	18.1
Total	13	38.1 (2.9)	267	810.3 (3.1)	5	30.0 (5.0)	36	323.0 (9.0)	10	341.0 (34.1)	20	312.0 (15.6)	348	1848.4
Test time	—	5400.0	—	4188.0	—	750.0	—	960.0	—	1620.0	—	1020.0	—	13938.0