Speech Attribute Detection to Recognize Arabic Broadcast Speech in Industrial Networks

Table 5

Best models of all Arabic attributes for both manner and place of articulation, with DBN-DNNs or CNNs with hidden layers from 1 up to 10 and 512 or 1024 units in each layer.

Attribute	CNN	CNN	DBN-DNN	DBN-DNN
Classes
Manner of articulation
Stop	12.22	9.14	13.11	12.34
Affricates	11.10	10.21	12.17	11.44
Fricatives	13.10	9.33	12.81	13.14
Nasals	13.14	9.80	12.62	13.11
Laterals	12.12	10.71	13.51	11.80
Trills	11.00	10.52	10.12	10.55
Approximants	11.31	9.11	12.33	11.61
Place of articulation
Labial	13.20	13.70	14.11	14.31
Labiodental	11.35	12.98	13.17	14.44
Interdental	11.42	12.11	14.81	14.86
Alveolar	12.12	13.97	14.36	14.50
Palatal	13.11	13.98	13.81	13.95
Velar	13.13	13.22	13.62	14.11
Uvular	12.61	12.66	14.62	14.22
Pharyngeal	11.51	12.71	13.62	13.55
Glottal	13.10	13.88	14.22	14.55