1	Comparative Study of Speech Intelligibility Inside Cars Angelo Farina¹, Fabio Bozzoli¹, Patricia Strasser²
2	Goals Evaluation of the acoustical confort inside a car, in terms of speech intellegibility Objective rating of both electroacoustical devices (sound system) and of natural communication between passengers Evaluation of the bi-directional performances of hands-free communication systems
3	Methods
4	The STI Method
5	MTF from Impulse Response It is possible to derive the MTF values from a single impulse response measurement:
6	Background noise If the background noise is superposed to the impulse response, the previous method already takes care of it, and the MTF values are measured correctly However, in some cases, it is advisable to perform a noise-free measurement of the IR, and then insert the effect of the noise with the following expression:
7	Transducers: binaural microphone A Cortex head and torso simulator was selected, after careful comparative tests performed in an anechoic chamber, which demonstrated its superiority to other binaural microphones (Neumann, B&K, Head Acoustics)
8	Transducers: mouth simulator The mouth simulator was built inside a ellipsoidal dummy head, employing low-cost parts. Its compliance with the ITU recommendation was confirmed by means of anechoic directivity tests.
9	Directivity measurements In both cases, the anechoic directivity measurements were performed employing a rotating table, directly synchronized with the sound board employed for measuring the impulse response. The Aurora software generates the required pulses on the right channel, which cause the rotating board to advance.
10	Directivity of the binaural microphone The simmetry revealed to be quite good, and the listening test of the sequence of impulse responses gives the impression of a pulsive source rotating around.
11	Directivity of the mouth simulator
12	Directivity of the mouth simulator
13	Directivity of a real human (I. Bork, PTB)
14	Frequency responses The binaural microphone exhibit the typical response of a dummy head, with significant boost around 4-5 kHz. The mouth simulator is flat between 200 and 1000 Hz, and requires substantial equalization outside this interval
15	Equalization of the mouth simulator The spectrum of the emitted test signal should correspond to the prescriptions of ITU T-P50 Recommendation. The overall SPL should be 67 dB(A) at 1m, on axis, for STI standard measurements
16	Equalization of the mouth simulator The MLS signal is prefiltered, so that the frequency response, measured at 1m in front of the mouth, complies with the IEC spectrum. The filtering is performed by means of the grahic equalizer incorporated in Cool Edit Pro.
17	Measurement example (no noise) The measured IR is saved as a TIM file, and processed with MLSSA
18	Measurement example (noise) The measured IR is saved as a TIM file, and processed with MLSSA
19	Verification of noise simulation The values of m(F) obtained by the measurement without noise were corrected for the S/N ratio, and compared with the m(F) values measured with noise
20	Experimental results on a test car The same car was measured at different speeds, employing two different software tools (DIRAC and MLSSA), and, with the latter, also with artificial noise compensation applied to the noise-free measurement.
21	Analysis of the results on a test car The measurements derived by artificial background noise correction of noise-free impulse responses revealed to provide, on average, slightly higher values of STI and much lower standard deviation of the results.
22	Conclusions The hardware and software developed allows for quick and reliable measurement of STI in cars. The background noise can be present during the actual measurement: however, it is possible to add its effect later, in two different ways: Mixing a noise recording over the re-recorded MLS signal, prior of IR deconvolution (yet to be assessed) Correcting the MTF values with the theoretical relationship, knowing the levels of the signal and of the noise (ideal method when only the noise spectral values are known, and no recording is available) The methodology developed, however, allows also for the creation of sound samples, containing speech (convolved with the noiseless IR) and background noise: these sound samples can be employed for listening tests.