1	Effects of the Background Noise on the Perceived Quality of Car Audio Systems Fabio Bozzoli, Enrico Armelloni, Emanuele Ugolotti, Angelo Farina
2	Goals Assess the quality of an automotive sound system by subjective listening test Analyze the effect of the background noise on the performances of the sound system Avoid the problems encountered when attempting to collect questionnaires from subjects during on-the-road tests
3	Methods
4	Binaural recordings
5	B-Format recordings
6	Binaural reproduction system
7	Dual-Stereo-Dipole vs. PanAmbio Dual Stereo Dipole 1 Binaural microphone (with pinnae) Different cross-talk cancelling filters for front and rear Stereo Dipoles, designed inverting the actual measurements taken on the dummy head placed in the listening room
8	Dual Stereo Dipole
9	DSP board for cross-talk cancellation
10	Algorithm for cross-talk cancellation The DSP board is simply used as a 2x2 convolver (FIR or WFIR filters) Thanks to the SIMD capabilities of the SHARC21161N processor, a single board can simultaneously perform the same processing (with different filtering coefficients) for the second Stereo Dipole The sets of filtering coefficients are computed by means of a modified version of the method developed by Nelson, Hamada and Kirkeby
11	Computation of the inverse filters The four inverse filters f are computed in the frequency domain, based on the measured head-related transfer functions h. The denominator, common to all the 4 filters, is a mixed-phase function Its inversion is possible introducing a small regularisation parameter e Making e variable with frequency, the inverse filters make optimal use of the limited number of taps available
12	Example of inverse filters
13	B-Format reproduction system
14	Bi-square Ambisonics array
15	DSP system for Ambisonics decoding
16	Algorithm for Ambisonics decoding Each speaker feed is basically simply a linear combination of the 4 inputs signals (WXYZ) The gains depend on the position of each particular loudspeakers A speaker-dependent FIR filter is added, for compensating its individual response curve, and providing some high-frequency phase randomization
17	Programming the decoder onto the SoundWeb
18	Effect of the FIR equalization of each loudspeaker
19	Complete system setup
20	ASK Listening room
21	Experimental results Background noise recordings on three very similar cars: - Opel Vectra 2.0 DTI (130 km/h) - Opel Zafira 2.0 DI (120 km/h) - Opel Zafira 2.0 DTI (120 km/h)
22	Software for automatic collection of questionnaires
23	Results The first results coming from the listening tests show that some light difference between the three cars can be systematically perceived (some subjects reliably identify each of the three cars, also after random shuffling the sound samples); Nevertheless, an ANOVA performed over the subjective responses, shows no significant difference among the three cars; The subjects employed for the tests revealed to be unsatisfactorily trained to listen to background noise; The scores obtained by the sound systems (IPA) are much worst in presence of the reproduced background noise than in absence of it;
24	IPA scores
25	Conclusions The sound quality evaluation is heavily affected by the presence of car’s background noise; The questionnaires employed for evaluation in absence of background noise revealed to be partially unsatisfactory for assessing sound systems in presence of background noise; Technicians used to evaluate sound systems by listening to music reproduction inside a silent environment need some further training for becoming used to listening with background noise; The hybrid reproduction systems revealed to be satisfactory for the reproduction of the sound recorded inside a car compartment, and can be further improved by means of a portable multichannel soundboard;