Distant-talking speaker identification by generalized spectral subtraction-based dereverberation and its efficient computation

EURASIP Journal on Audio, Speech, and Music Processing

Table 9 Speaker recognition rates in real environment

Method	Speaker	Processing at	Recognition
number	models	test stage	rate (%)
1 (Baseline)	Clean speech	CMN with	61.5
	models
		beamforming
2	Reverberant	CMN with	72.5
	speech models	beamforming
3 (LTLSS-based	Dereverberant	LTLSS-based	81.0
method)	speech models
	based on LTLSS	dereverberation
4 (MSLP-GSS-based	Dereverberant	MSLP-GSS-based	83.8
method)	speech models
	based on	dereverberation
	MSLP-GSS
5 (MCLMS-GSS-based	Dereverberant	MCLMS-GSS-based	87.8
method)	speech models
	based on	dereverberation
	MCLMS-GSS