An Adaptive Method Based on Multiscale Dilated Convolutional Network for Binaural Speech Source Localization

Table 2

Localization accuracy (%) of different approaches in additively noisy environments.

SNR	—	−10 dB			0 dB			10 dB			20 dB			30 dB
Noise	Avg.	White	F16	M109	White	F16	M109	White	F16	M109	White	F16	M109	White	F16	M109
MLP [8]	83.77	62.93	53.21	67.41	72.65	71.37	82.37	81.62	86.75	95.51	89.42	96.26	99.15	98.18	99.89	99.89
DNN [19]	82.56	43.16	38.25	53.42	70.30	57.16	86.32	97.65	92.31	100.0	99.89	99.89	100.0	100.0	100.0	100.0
Regular CNN	84.65	54.38	41.99	65.60	73.61	69.76	86.43	89.96	90.81	99.04	98.72	99.79	99.79	99.89	100.0	100.0
Dilation-2 CNN	87.46	45.30	54.17	75.75	70.94	77.67	97.76	97.33	93.59	99.68	99.89	99.89	100.0	100.0	100.0	100.0
Dilation-5 CNN	90.14	62.61	54.17	80.34	83.55	75.85	99.15	97.86	98.61	100.0	100.0	100.0	100.0	100.0	100.0	100.0
Cascaded DCNN	89.62	57.05	54.38	87.61	76.92	76.50	99.25	95.51	97.22	100.0	99.79	100.0	100.0	100.0	100.0	100.0
Ours	89.34	59.83	47.54	78.63	84.19	74.47	98.61	98.40	98.50	100.0	100.0	100.0	100.0	100.0	100.0	100.0
Ours	91.85	68.16	56.62	90.06	86.00	80.24	99.36	98.61	98.72	100.0	100.0	100.0	100.0	100.0	100.0	100.0