Wideband Hands-Free in Cars – New Challenges for System Design and Testing - Digital Signal Processing for In-Vehicle Systems and Safety - page 106

Fig. 6.2 Users' satisfaction depending on delay and TELR (TELR ¼ SLR + RLR + Echo Loss)

from [ 5 ]

6.3 Listening Speech Quality

The performance requirements for the speech quality in receiving are probably
easiest to fulfill due to the high quality of built-in car audio systems. For aftermarket
hands-free systems, this is much more challenging. The extension of the frequency
range in sending direction not only provides better representation of the low-
frequency components of the transmitted speech but also increases the amount of
noise transmitted by the microphone. This is of particular importance because the
in-car noise is dominant in the low-frequency range. It imposes additional quality
requirements on all speech enhancement techniques such as beamforming for
microphones, noise cancelation, and others.
An objective measure 3QUEST according to ETSI EG 201 396-3 [ 6 ] is capable
of determining the speech, noise, and overall quality, and such can be used in the
optimization of wideband hands-free systems. The algorithm calculates correlation
between the processed signal - typically recorded in sending direction of a hands-free
system (uplink) - and two references, the original clean speech signal and the signal
recorded close to the hands-free microphone. This signal consists of the near-end
speech and the overlapped in-car noise. The algorithm is described in [ 6 ]and[ 7 ]
in detail. Statistical analyses lead to a one-dimensional speech quality score (S-MOS),
a noise quality score (N-MOS), and an overall quality score representing the general
impression (G-MOS). The algorithm is narrowband and wideband capable and
provides correlations in the range of > 0.91 to the results of subjective tests.