EURASIP Journal on Audio, Speech, and Music Processing

Table 13 Top 10 positive distractor events for music (event labels related to false positive decisions of the network about the “music” class)

From: Exploring convolutional, recurrent, and hybrid deep neural networks for speech and music detection in a large audio dataset

Event	Event ID	Ratio	\(d^{+}_{mu}\)
Percussion	/m/0l14md	110/137	0.600
Pizzicato	/m/0d8_n	73/78	0.588
Drum	/m/026t6	82/94	0.585
Organ	/m/013y1f	74/81	0.582
Keyboard (musical)	/m/05148p4	77/87	0.578
Brass instrument	/m/01kcd	85/116	0.524
Singing	/m/015lz1	51/62	0.471
Hammond organ	/m/03gvt	43/46	0.466
Bass drum	/m/0bm02	43/47	0.461
Tabla	/m/01p970	40/41	0.459

\(d^{+}_{mu}\) score (Eq. 15) is used to rank the events. The ratio column shows the number of false positives for music where the distractor event label is found (numerator) and the number of non-music segments that contain the distractor event (denominator)

Back to article page