Instrumental Estimation of E-Model Parameters for Wideband Speech Codecs

EURASIP Journal on Audio, Speech, and Music Processing

Table 2 Pearson correlation coefficients and root mean square prediction errors between the auditory and the estimated MOS for Databases 1–5 and 7, using different WB signal-based models. These measures are computed after third-order mapping.

Database		No. of stimuli		WB-PESQ		TOSQA-2001		Mod. WB-PESQ
		NB	WB
1		100	44	0.93	0.36	0.91	0.40	0.93	0.34
		NB only		0.97	0.29	0.90	0.41	0.98	0.28
2		36	36	0.83	0.49	0.78	0.55	0.83	0.49
		NB only		0.75	0.47	0.77	0.44	0.65	0.49
3		112	392	0.97	0.23	0.92	0.35	0.96	0.24
		NB only		0.94	0.24	0.69	0.37	0.95	0.19
4		96	1288	0.90	0.29	0.88	0.31	0.92	0.25
		NB only		0.88	0.43	0.83	0.41	0.90	0.25
5		360	360	0.93	0.35	0.88	0.46	0.97	0.23
		NB only		0.93	0.37	0.78	0.48	0.95	0.26
7	AA	0	336	0.94	0.30
	AC	0	336	0.95	0.30
	BB	0	336	0.97	0.19
	BD	0	336	0.97	0.19
	CC	0	336	0.93	0.33
	CE	0	336	0.89	0.44
	DD	0	336	0.97	0.18
	DA	0	336	0.97	0.19
	EE	0	336	0.83	0.56
	EB	0	336	0.87	0.51

Database 7: Letters specify the combination of the test laboratory (A–E, first letter) and the tested candidate codec (A–E, second letter).