MUS_TECH 348
3-D Sound and Spatial Audio
Wightman & Kistler (1989)
Headphone simulation of free-field listening
I. Stimulus synthesis
II. Psychophysical validation
I. Stimulus synthesis
Goal is to be able to capture free-field listening
acoustics with headphones.
• 200-14,000 Hz
• Greater than 20 dB S/N (only 20 dB?)
• 8 loudspeakers on movable arch creating 144 directions
• With & without bite bar
Measure loudspeaker-delivered HRTFs and compare to
headphone-delivered HRTFs
HRTF
measurement
system
Left ear
Variability in
HRTF
measurements
Assembly replaced
10 times with bite bar
Right ear
Left ear
Assembly left in place
with no bite bar
Right ear
Headphone
replacement with
assembly in place
HRTF intersubject variability
II. Psychophysical validation
Goal is to compare localization performance in freefield and headphone listening
Stimuli:
8 250 msec noise bursts
200 - 14,000 Hz
random spectral changes by critical band
Presentation:
6 loudspeakers at a time mounted on arch
headphones
72 positions
Task:
absolute judgment of azimuth and elevation
no measure of distance or quality
Types of Errors
•Angle error (mean of difference angles)
•Judgment centroid (average direction)
•Dispersion of judgments
•Front-back reversals are removed! (and examined
separately)
Results
•Substantial individual differences
•Less obvious in global measures
•Most evident in elevation judgments
•Performance varies with region
•Best localization: side (contradicts other studies)
•Worst localization: top rear
•Free-field and headphone judgments very similar
•More front-back reversals with headphones
Headphone simulation data in parentheses
SDE has most errors.
SDO has fewest errors, especially for elevation.
Elevation Dependency
Function
Interaural intensity difference
compared to 0-degrees elevation
Subject SDE’s poor elevation
judgments could be explained by the
lack of a coherent pattern
Begault: Challenges to the Successful
Implementation of 3-D Sound
Focus is on deployable systems, especially audio systems
Individual HRTFs can be quite different
Challenges:
Eliminate front-back reversals & improve externalization
Reduce HRTF data load
Resolve conflicts in data specifications
Begault: Challenges to the Successful
Implementation of 3-D Sound
Mismatch of Specification and Performance
Success depends on:
HRTFs:
some work better than others
different sets create timbral percepts
Input sounds
broadband sounds localize better
Specification
Have reasonable expectations
What kinds of HRTFs to use for systems?
General HRTFs designed for average listeners
HRTFs of good localizer
Reality vs
Ideal
From Begault and Wenzel, 1993
Begault: Challenges to the Successful
Implementation of 3-D Sound
Localization error
For dummyhead recordings, 30% of locations
suffer reversals
4:1 front-back vs back-front
Many sounds not externalized
Low-frequency Response Errors
Measurement equipment can’t get it right
Data-reduction for HRTFs
Reduce the number of coefficients
Alternative Strategies like pole-zero modeling
Martens: Perceptual evaluation of filters
controlling source direction: Customized
and generalized HRTFs for binaural
synthesis
Focus is on systems supporting directional
hearing with special consideration on HRTF
design
Position of sound source and position of auditory
event do not always coincide, but that is not
necessarily an issue of accuracy
Sound localization might better be called space
perception
Martens: Perceptual evaluation of filters controlling source
direction: Customized and generalized HRTFs for binaural
synthesis
Binaural Synthesis
Good localizer HRTFs not supported by evidence
Given the variety of approaches to binaural synthesis, better to use
the term Directional Transfer Functions (DTFs) when they are created
analytically
Target
One
Many
Exact
Individualized HRTFs
Averaged HRTFs
Analytic
Customized DTFs
Generalized DTFs
Performance evaluation (in additional azimuth and elevation);
Externalization
Range
Coherence
Naturalness
Martens: Perceptual evaluation of filters controlling source
direction: Customized and generalized HRTFs for binaural
synthesis
Binaural Synthesis Evaluation
What features are needed to make binaural synthesis “ear adequate”
Binaural cues can be based on analysis and selected resynthesis
Principle Components Analysis (PCA)
Selective Reconstruction (for example, leaving out phase
information
[Pole-zero design]
Elevation judgments needed only three out of four cues:
ipsilateral magnitude
interaural magnitude
ipsilateral phase
interaural phase
Martens: Perceptual evaluation of filters controlling source
direction: Customized and generalized HRTFs for binaural
synthesis
Customizing HRTFs
Calibration methods:
Anthropometric (anatomy)
Acoustic (HRTFs)
Psychophysical (perception)
Source Range
Ipsilateral gain and contralateral attenuation are
important