Mid-Term ReviewMid-Term Review
John W. WorleyJohn W. Worley
AudioGroup, WCLAudioGroup, WCL
Department of Electrical and Computer EngineeringDepartment of Electrical and Computer Engineering
University of Patras, GreeceUniversity of Patras, Greece
http://www.wcl.ee.upatras.gr/AudioGroup/http://www.wcl.ee.upatras.gr/AudioGroup/
Slide 2 of 17
TasksTasks
• 2.1 The precedence effect • Franssen illusion
• 2.2 Reliability of auditory cues in multi-source scenarios • Learning non-individualised HRTFs
• 2.3 Perceptual models of room reverberation with application to speech recognition • Complex smoothed room responses• Perceptual factors in room responses
Slide 3 of 17
Task 2.1Task 2.1 Franssen illusion Franssen illusion
• Reverberant environments = cue to multiple directions.
• The precedence effect = stable directional percept.
• Franssen illusion (F.I.)• Precedence effect.• ITD/ILD dependant
Slide 4 of 17
Task 2.1Task 2.1 Franssen illusion Franssen illusion
Hypothesis• Localisation requires transients.• Signal spectral density.• Room differences.• ITD/ILD dependant.
Solution• Various onset transitions.• Sinusoid & Harmonic complex’s.• Large vs. small rooms
At present: • F.I. in reverberation chamber.• No transition effect.• Increasing spectral density = Increased localisability.
• F.I. dependant on poor stimuli localisability.
Future: • F.I. with Grouping cues??
Slide 5 of 17
Task 2.2Task 2.2 Learning non-individualised HRTFs Learning non-individualised HRTFs
Cone-of-confusion
MVP HRTFs
Individual HRTFs
Slide 6 of 17
Task 2.2Task 2.2 Learning non-individualised HRTFs: Results Learning non-individualised HRTFs: Results
Type – I (2 listeners) Type - II (3 listeners)
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10Day
% R
ever
sals
0
20
40
60
80
100
1 2 3 4 5 6 7 8 9 10Day
% R
ever
sals
Back-Front
Front-Back
Total
• Response bias significantly determines reversal type
= No reversal predisposition. = Majority of front-to-back reversals.
Slide 7 of 17
Task 2.3 Task 2.3 Complex Smoothing Room Impulse Response (RIR):Complex Smoothing Room Impulse Response (RIR):
time domain frequency domain
Original
RIR
Smoothed
RIR
Slide 8 of 17
perceptual smoothing profilesperceptual smoothing profiles
Start with a “smoothed” room response
Use smoothing based on perceptionvariable spectral resolutionvariable spectral resolutionvariable frequency-dependent windowingvariable frequency-dependent windowing
Employ “room masking models”
Slide 9 of 17
Task 2.3 Task 2.3 Inverse filtering using smoothed filtersInverse filtering using smoothed filters
0 5 10 15 20
-80
-60
-40
-20
0
Original
Tim
e E
nerg
y (d
B)
Time (msec)
Complex Smoothed
10 100 1k 10k-40
-30
-20
-10
0
10
20
30
OriginalMag
nitu
de (
dB)
log Frequency (Hz)
Complex Smoothed
100 105 110 115 120
-80
-60
-40
-20
0
Original Equalised
Tim
e E
nerg
y (d
B)
Time (msec)10 100 1k 10k
-40
-30
-20
-10
0
10
20
30
Original Equalised
Mag
nitu
de (
dB)
log Frequency (Hz)
from: “Results for Room Acoustics Equalisation Based on Smoothed Responses”Panagiotis D. Hatziantoniou and John N. Mourjopoulos,114th AES Convention, Amsterdam, March 2003
time domain frequency domain
modification
compensation
Slide 10 of 17
Tests in 6 rooms of Volume 60m3 – 11000m3
EDT reduced by up to 0,5 sec
C80 improves by up to 5 dB
D50 improves by up to 20%
Spectral deviation is reduced up to 4 dB
Task 2.3 Task 2.3 Smoothed filters physical metricsSmoothed filters physical metrics
from: “Results for Room Acoustics Equalisation Based on Smoothed Responses”Panagiotis D. Hatziantoniou and John N. Mourjopoulos, 114th AES Convention, Amsterdam, March 2003
Slide 11 of 17
Task 2.3 Task 2.3 Perceptual factors in room responsesPerceptual factors in room responses
• Real-time perception test.• Various stimuli types (steady-state & transients).• Assess multiple perceptual factors.
Slide 12 of 17
Task 2.3 Task 2.3 Perceptual factors in room responsesPerceptual factors in room responses
• Source width.• Source
distance.• Envelopment.
Slide 13 of 17
Task 2.3 Task 2.3 Perceptual factors in room responsesPerceptual factors in room responses
• Anchor end-points with illustrative demonstrations and explanation.
• Results subjected to factor analysis
Slide 14 of 17
Future workFuture work
• Perceptual factors in room responses (2.3).
• ITD/ILD plausibility cues (2.1, 2.2).
• The combination of the cues is still debated.
• Use F0 grouping with FI for hierarchy of
cues (2.2).
AudioGroup, WCLAudioGroup, WCLDepartment of Electrical and Computer EngineeringDepartment of Electrical and Computer EngineeringUniversity of Patras, GreeceUniversity of Patras, Greecehttp://www.wcl.ee.upatras.gr/AudioGroup/http://www.wcl.ee.upatras.gr/AudioGroup/