Spatial speech detection for binaural hearing aids using deep phoneme classifiers

Topical Issue - Auditory models: from binaural processing to multimodal cognition

Open Access

Table 1

Speech source localization performance in hit rate (Eq. (16)) for the diffuse noise scenarios obtained with all different localization approaches – spatial probability only (DOA, Eq. (14)), speech quality measure only (ASQM, Eq. (15)), and the combination of DOA and ASQM (DOA ASQM, Eq. (13)) – and ASR-based speech quality measures: M-Measure (MM), Matched Phoneme (MaP) filters, and Inverse entropy (iEnt). Results are shown for the anechoic (top panel) and the reverberant office (bottom panel) environment depending on the SNR, the best performing approach per SNR is highlighted in bold.

Anechoic
	DOA	ASQM			DOA × ASQM

SNR		MM	MaP	iEnt	MM	MaP	iEnt
−10	100	97	13	0	100	95	34
−5	100	99	45	0	100	100	69
0	100	85	34	0	100	100	92
5	100	57	25	19	100	100	99
10	100	31	27	16	100	100	100
15	100	20	17	78	100	99	100
20	100	12	11	44	100	98	100
Office
	DOA	ASQM			DOA × ASQM

SNR		MM	MaP	iEnt	MM	MaP	iEnt

−10	97	35	14	0	96	76	3
−5	100	35	19	0	100	89	27
0	100	30	37	0	100	93	60
5	100	25	38	3	100	100	95
10	100	11	26	6	100	97	100
15	100	7	20	34	100	92	100
20	100	3	6	23	100	89	100

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.