Capturing the elevation dependence of ITD using an extension of the - - PowerPoint PPT Presentation
Capturing the elevation dependence of ITD using an extension of the spherical-head model Rahulram Sridhar 3D3A Laboratory, Princeton University This talk is based on the work by Sridhar and Choueiri, published under the same title as paper
Rahulram Sridhar 3D3A Laboratory, Princeton University
This talk is based on the work by Sridhar and Choueiri, published under the same title as paper number 9447 in the proceedings of the AES 139th convention
Top view of human head
sound source
Top view of two different human heads showing variations in head shape and size
Images taken from Ball et al. “A comparison between Chinese and Caucasian head shapes,” Applied Ergonomics 41 (2010) 832-839
From measured HRIRs3,5
Techniques for estimating individualized ITDs
3D scan + acoustic ray tracing6 Anthropometry-based model based on spatial variation of ITD7-10 Anthropometry-based model based on simple geometric shape of head4,12
References correspond to those in paper 9447 on which this talk is based
From measured HRIRs3,5
Techniques for estimating individualized ITDs
Numbered references correspond to those in paper 9447 on which this talk is based
Sound localization facility at Wright Patterson Air Force Base in Dayton, Ohio
Image taken from Hartmann, W. M., “How We Localize Sound,” Physics Today, pp. 24-29, 1999
Techniques for estimating individualized ITDs
3D scan + acoustic ray tracing6
Propagation delay visualization. Sound source is 39 deg. to the right of the dummy head
Image take from Gamper, H. et al., “Estimation of multi path propagation delays and interaural time differences from 3-D head scans,” ICASSP 2015
Numbered references correspond to those in paper 9447 on which this talk is based
Techniques for estimating individualized ITDs
Anthropometry-based model based on simple geometric shape of head4,12
References correspond to those in paper 9447 on which this talk is based
Techniques for estimating individualized ITDs
Anthropometry-based model based on simple geometric shape of head4,12
References correspond to those in paper 9447 on which this talk is based
Ellipsoidal-head model4 fairly simple and accurate
Techniques for estimating individualized ITDs
Anthropometry-based model based on simple geometric shape of head4,12
References correspond to those in paper 9447 on which this talk is based
Ellipsoidal-head model4 fairly simple and accurate tricky mapping to anthropometric features
Techniques for estimating individualized ITDs
Anthropometry-based model based on simple geometric shape of head4,12
References correspond to those in paper 9447 on which this talk is based
Ellipsoidal-head model4 Spherical-head model12 fairly simple and accurate tricky mapping to anthropometric features
Techniques for estimating individualized ITDs
Anthropometry-based model based on simple geometric shape of head4,12
References correspond to those in paper 9447 on which this talk is based
Ellipsoidal-head model4 Spherical-head model12 fairly simple and accurate tricky mapping to anthropometric features easy mapping to anthropometric features
Techniques for estimating individualized ITDs
Anthropometry-based model based on simple geometric shape of head4,12
References correspond to those in paper 9447 on which this talk is based
Ellipsoidal-head model4 Spherical-head model12 fairly simple and accurate tricky mapping to anthropometric features easy mapping to anthropometric features not so accurate
Techniques for estimating individualized ITDs
Anthropometry-based model based on spatial variation of ITD7-10
References correspond to those in paper 9447 on which this talk is based
References correspond to those in paper 9447 on which this talk is based
Woodworth & Schlosberg formula11 Introduction
References correspond to those in paper 9447 on which this talk is based
Spherical-head (Top View)
sound source
is speed of sound
Model formulation
Model formulation
Fit single optimized head radius per listener for many listeners
Model formulation
measured anthropometric data
width depth height
Fit single optimized head radius per listener for many listeners
Model formulation
measured anthropometric data
width depth height
Fit Fit single optimized head radius per listener for many listeners
Model formulation
measured anthropometric data
width depth height
Fit Fit single optimized head radius per listener for many listeners individualized
substitute for
Limitations
▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲
45 90 135 180 325 350 375 400 425 450 Elevation (deg.) Mean ITD (μs)
Azimuth = 45°
Measured
Limitations
▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲
90 135 180 325 350 375 400 425 450 Elevation (deg.) Mean ITD (μs)
Azimuth = 45°
Measured Spherical-head model
Limitations
Woodworth & Schlosberg formula11 Introduction
References correspond to those in paper 9447 on which this talk is based
Spherical-head (Top View)
sound source
Woodworth & Schlosberg formula11 Introduction Formula for estimating low-frequency ITD1
References correspond to those in paper 9447 on which this talk is based
Spherical-head (Top View)
sound source
Introduction
Interaural coordinate system azimuth elevation
interaural axis frontal axis
half-plane
Model formulation
Model formulation
Fit single optimized head radius per listener per half-plane for many listeners
performed per half-plane
Model formulation
Fit single optimized head radius per listener per half-plane for many listeners
performed per half-plane
measured anthropometric data
width depth height arc length (approx.)
Model formulation
Fit single optimized head radius per listener per half-plane for many listeners
performed per half-plane
measured anthropometric data
width depth height arc length (approx.)
Fit
performed per half-plane
Model formulation
Fit single optimized head radius per listener per half-plane for many listeners
performed per half-plane
measured anthropometric data
width depth height arc length (approx.)
Fit individualized
substitute for performed per half-plane
Model formulation (low-frequency ITD)
Fit single optimized head radius per listener per half-plane for many listeners
performed per half-plane
measured anthropometric data
width depth height arc length (approx.)
Fit individualized
substitute for performed per half-plane
High-frequency ITD estimation accuracy
40 80 20 40 60 Azimuth (deg.) Mean Absolute ITD Error (μs)
Spherical-head model
High-frequency ITDs
High-frequency ITD estimation accuracy
■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■
40 80 20 40 60 Azimuth (deg.) Mean Absolute ITD Error (μs)
Spherical-head model Proposed model Points offset horizontally by ±0.5° for clarity
High-frequency ITDs
Low-frequency ITD estimation accuracy
40 80 80 160 240 Azimuth (deg.) Mean Absolute ITD Error (μs)
Spherical-head model
Low-frequency ITDs
Low-frequency ITD estimation accuracy
■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■
40 80 80 160 240 Azimuth (deg.) Mean Absolute ITD Error (μs)
Spherical-head model Proposed model
Low-frequency ITDs
Ability of proposed model to capture elevation
dependence of measured ITDs
▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲
90 135 180 325 350 375 400 425 450 Elevation (deg.) Mean ITD (μs)
Azimuth = 45°
Measured Spherical-head model
Ability of proposed model to capture elevation
dependence of measured ITDs
▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲ ▲
■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■
45 90 135 180 325 350 375 400 425 450 Elevation (deg.) Mean ITD (μs)
Azimuth = 45°
Measured Spherical-head model Proposed model
High-frequency ITDs
Contribution of anthropometric features to head radius
■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■
45 90 135 180 2 4 6 Elevation (deg.) Contribution to spherical-head radius (cm)
Head width
High-frequency ITD
Contribution of anthropometric features to head radius
■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■
90 135 180 2 4 6 Elevation (deg.) Contribution to spherical-head radius (cm)
Head width Head depth
High-frequency ITD
Contribution of anthropometric features to head radius
■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■
45 90 135 180 2 4 6 Elevation (deg.) Contribution to spherical-head radius (cm)
Head width Head depth Pinnae mean arc length
High-frequency ITD
Contribution of anthropometric features to head radius
■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■
■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■
45 90 135 180 2 4 6 Elevation (deg.) Contribution to spherical-head radius (cm)
Head width Head depth Pinnae mean arc length Head width
High-frequency ITD Low-frequency ITD
Contribution of anthropometric features to head radius
■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■
■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■ ■
90 135 180 2 4 6 Elevation (deg.) Contribution to spherical-head radius (cm)
Head width Head depth Pinnae mean arc length Head width Head depth
High-frequency ITD Low-frequency ITD
dependence of ITDs
azimuths
captures elevation dependence of ITDs
when estimating high- (low-) frequency ITDs.
estimate varies based on sound source location
frequency ITD for sources behind the listener
sources in front of the listener