1. Sound Quality Assessment IR measurements and AQT
University
of Parma
Sound Quality Assessment IR measurements and AQT

2. Problems with traditional measurements methods
Traditional measurement are based on steady-state signals - they provide a frequency response curve - this does not provide significant match with listening experience in cars
It was attempted to employ transient analysis (impulse response) based on the same parameters employed for concert halls (reverberation time, clarity, definition) - also these do not correspond with the short time scale of time transients inside cars
In-car listening tests are useful, but difficult to standardize
General lack of understanding of the relationship between subjective perception and physical facts

3. Subjective – Objective discrepancy1
Traditional measurements: steady-state frequency response, impulse response 2
In-car (simple blind) listening tests 3
Off-car (double blind) listening tests based on binaural or B-format recordings
GOALS
Assessment of the absolute quality of a car sound system
Comparative ranking of different sound systems
PROBLEM
The three methods give, in general, different results
SOLUTIONS
Advanced measurement techniques (Sweep+AQT),
self-stabilizing subjective questionnaires (IPA)

4. Solutions
Reliable in-car measurement technique (ESS, exponential sine sweep)
Transient analysis and equalization (AQT)
Virtual listening for comparative tests
Subjective (IPA) and objective (IPM) evaluation indexes for audio quality assessment
Usage of the AQT spectrum for designing subjectively-optimized equalizing filters

5. Methods: Reliable in-car measurement technique
Impulse Response (IR) as standard measurement for characterization of the car sound
Hardware: omnidirectional microphone mounted on a torso
Hardware: Notebook PC + Edirol FA-101 Firewire interface
Software: Adobe Audition and Aurora for IR measure
AQT as advanced analysis (instead of FFT)

6. Standardized Position of the Microphone
Reliable in-car measurement technique
Standardized Position of the Microphone
ca. 25o
70 cm
30 cm
verticale volante
verticale a piombo
asse del dorso (sul piano dorsale)
tettuccio
After years of tests with binaural microphones, we concluded that for a reliable evaluation of the frequency response of the sound system it is better to employ a standard omnidirectional microphone, but it needs to be mounted on a torso

7. Hardware: PC and audio interface
Reliable in-car measurement technique
Hardware: PC and audio interface
Edirol FA-101
Firewire sound card:
10 in / 10 out
24 bit, 192 kHz
ASIO and WMA

8. Software Reliable in-car measurement technique Aurora Plugins
Generate MLS
Deconvolve MLS
Generate Sweep
Deconvolve Sweep
Convolution
Kirkeby Inverse Filter
Speech Transm. Index

9. Measurement process
The desidered result is the linear impulse response of the acoustic propagation h(t). It can be recovered by knowing the test signal x(t) and the measured system output y(t). It is necessary to exclude the effect of the not-linear part K and of the background noise n(t).

10. Test signal: Exponential Sine Sweep
x(t) is a sine signal, which frequency is varied exponentially with time, starting at f1 and ending at f2.

11. Test Signal – x(t)
Stop

12. Measured signal - y(t) Stop
The not-linear behaviour of the loudspeaker causes many harmonics to appear

13. Inverse Filter – z(t) Stop
The deconvolution of the IR is obtained convolving the measured signal y(t) with the inverse filter z(t) [equalized, time-reversed x(t)]

14. Equalization of the inverse filter
The inverse filter is the “Time reversal mirror” of the test signal, but the spectrum must be adjusted, as the original signal had a spectrum falling down by 3 dB/octave.
test signal x(t)
Inverse Filter z(t)

15. Deconvolution of Log Sine Sweep
The “time reversal mirror” technique is employed: the system’s impulse response is obtained by convolving the measured signal y(t) with the time-reversal of the test signal x(-t). As the log sine sweep does not have a “white” spectrum, proper equalization is required
Test Signal x(t)
Inverse Filter z(t)

16. Deconvolution = rotation of the sonogram
2nd order
Linear
Convolution with inverse filter rotates the time-frequency plane counter-clockwise

17. Result of the deconvolution1° 2° 3° 5°
The last impulse response is the linear one, the preceding are the harmonics distortion products of various orders

18. IR Selection
After the sequence of impulse responses has been obtained, it is possible to select and insulate just one of them:

19. Maximum Lenght Sequence vs. Sweep

20. Post processing of impulse responses
Specific XFM modules have been developed for the computation of STI and for performing standard analysis according to ISO-3382

21. AQT Analysis
Goal:
Substitute classical analysis based on FFT with a new kind of dynamic transient analysis
Reason:
FFT characterizes a only in steady state condition
BUT
Hearing is more sensible to transients
Musical signal is not a steady state signal
Solution:
AQT analysis (Acoustic Quality Test)

22. AQT Analysis Features:
Use transient stimulus (wavelet) instead of steady-state (periodic sine)
Find a frequency response curve more related to perceived curve than FFT response
Find resonances and tails that affect hearing
Useful during car development and for design of optimal equalization
Output:
“Dynamic” frequency response curve of the sound
Articulation curve (transient capability at each frequency)

23. AQT Analysis
Methods:
The test signal is a sequence of sine bursts of increasing frequency
Each burst is 200 ms long. The gap between a burst and the next is 33 ms
The instantaneous RMS level is plotted versus time (or frequency)
The same result can be also obtained convolving the test signal with the IR

24. AQT Analysis
Details:
The envelope of the maxima is the dynamic frequency response of the
The depth of the valley between two bursts shows the dynamic transient capability of the
Universiy of Parma

25. AQT Analysis Temporal masking according to Zwicker
The analysis of a single burst shows the level variation due to a single burst, and the overshoot phenomenon at the beginning and end of the burst.

26. AQT busrt shaping
Abrupt signal attack causes large bandwidth during attack and release
The burst is multiplied with an exponential envelope to control bandwidth
The spectrogram shows the behaviour changes
amplitude
time
amplitude
time
frequency
time

27. AQT phase
Mixed approach: different calculation method for different frequency ranges: FFT phase, burst phase, minimum phase
Burst delay: the phase is computed from the time delay between the burst and the burst response
■ Burst
■ Burst response
amplitude
time
Phases are mixed together
Magnitude
Magnitude
Phase
Phase

28. Equalization Filter Synthesis
Many IRs measured inside the car cockpit, the most part is located near to the listening positions
Kirkeby inverse filtering method
Regularization parameter
More microphones than speakers
Linear algebra over determined problem
Least square solution

29. Pre-ringing
Sometimes inverse filters contains pre-ringing that cause audible artifacts
Solution: linear-phase low-pass filter with time-variant cut-off frequency
The time variant filtering has negligible effects on magnitude and phase
Result: a single sided IR without audible artifacts.

30. AQT Analysis
AQT Tool:
An automatic tool was developed to perform a quick processing
This tool can also generate the coefficients for digital FIR filters
The filters are designed for producing a prescribed optimal frequency response

31. AQT Equalization results
Fiat STILO
Above: measured AQT curve not equalized vs. target curve (red)
Below: measured AQT curve after equalization vs. target curve (red)
Results (IPA listening test):
Not equalized
Traditional FIR
Inverse AQT
7.18
7.20
7.50

32. The new AQT Matlab program
The new module is still under development and will allow for very fast computation of the AQT curve and of the equalizing filter coefficients

33. Conclusions
The sine sweep method revealed to be systematically superior to the MLS method for measuring electroacoustical impulse responses
In fact, it is now employed from top-grade measurement systems, includiong the Audio Precision (TM)
Traditional FFT-based analysis of the measured impulse response did noy show good correlation with subjective perception
It was tehrefore necessary to derive new physical descriptors, better correlating with the human listening experience inside a car.
The AQT spectrum revealed to be a much better descriptor of the spectrum prceived by humans than the FFT spectrum
Equalizers designed taking into account the AQT spectrum outperform other equalization strategies