Wideband Hands-Free in Cars – New Challenges for System Design and Testing - Digital Signal Processing for In-Vehicle Systems and Safety - page 113

Fig. 6.12 3D Relative Approach RAe-r(t,f) between the echo signal e and the reference ear

D D

signal r

The parameters echo loss, echo delay, and m D RAe-r are used as input signal for
a linear regression in order to correlate the objective results to the subjective MOS
for the echo model.
In the first step, only the two parameters echo loss and echo delay were used in the
regression. The result is shown in the left-hand scatterplot in Fig. 6.13 . A correlation
of r 0.80 is achieved, but the comparison of auditory MOS and objective MOS

shows systematical errors: clusters of identical objective MOS occur in Fig. 6.13
(see arrows), which spread over a wide range of auditory MOS (between approxi-
mately 1.7 and 3.7 MOS). This can be explained by the different spectral content of
these echo signals leading to significant different echo ratings in subjective tests -
although the objective parameters (echo delay, echo attenuation) are identical.
The plot on the right-hand side in Fig. 6.13 shows the correlation between the
auditory MOS and the objective results based only on the two-dimensional mean
value m

¼

0.84. The systematical error
is implicitly solved using the Relative Approach-based analysis. In principal, this
could be expected because the Relative Approach considers the sensitivity of
human hearing, especially for different frequency characteristics of transmitted
sounds.
The combination of the three parameters m

RAe-r . The correlation factor increases to r ¼

D

RAe-r , echo loss, and echo delay to
the objective MOS further increases the correlation ( r

D

¼

0.90). The scatterplot is
shown in Fig. 6.14 (left-hand side) together with the error distribution in the right-
hand picture. The residual error between objective and auditory MOS is below 0.5
MOS in 84% of test conditions.