EURASIP Journal on Audio, Speech, and Music Processing

Table 16 Upper bound and baseline for the experiments using the real reverberant data

From: Single-channel dereverberation by feature mapping using cascade neural networks for robust distant speaker identification and speech recognition

Method	Acoustic model	Digit accuracy (%)
Method	Acoustic model	InCar	Lounge	MeetR	Office	Average
Close (ideal)	Clean	97.3	96.5	97.3	98.1	97.3
Close (ideal)	Multi	90.0	88.7	91.6	89.1	89.9
Baseline	Clean	76.4	43.8	89.2	85.1	73.6
Baseline	Multi	82.9	74.3	93.3	85.6	84.0
CMN	Clean	84.9	59.8	90.7	93.5	82.2
CMN	Multi	80.1	83.1	89.9	84.4	84.4

Back to article page