R T I D

R T I D

5. Anthropometry-based ITD individualization

5. Anthropometry-based ITD individualization

The next step after designing a software capable of real-time ITD scaling is to provide a method that, based on anthropometric measures, scales the ITD originally contained in the

BRIRs (i.e. that of the HATS FABIAN) that accomplishes the individualization task. The intrinsic individual characteristics of the ITD were covered in chapters

1

and

2 . It was

explained that the morphology of the subject plays a decisive role in the constitution of interaural delays.

Therefore, this chapter deals with the development of an empiric formula to find the individual ITD scaling factor based on anthropometric measures.

5.1. Listening test: ITD Individualization by scaling

The ITD individualization software was used in a listening test where the ITD of a foreign data set should be individually scaled by different subjects. Similar to the work of

Algazi et al.

( 2001b

) for finding an optimal head radius (see sec.

2.3

), relevant anthropomet-

ric measures: head height, head width and head depth according to the norm

DIN33402-2E

( 2005

) of body dimensions as well as a new measure the intertragus distance were considered for finding an individual ITD scaling factor by means of multiple regression.

The last measure, the intertragus distance, represents the distance between the incisura

anterior of left and right ears marking the tragus upper end. This measure was included because of its proximity to the ear-canal and the ease of its determination. Figure

5.1

shows the above mentioned anthropometric measures.

5.1.1. Test setup

A binaural data set of BRIRs was recorded in a damped room (Volume=155 m

3

, RT = 0.47

s) with 1

◦ azimuthal resolution, covering 180

(+-90

) of the frontal plane not including elevation. The HATS FABIAN (Fast and Automatic Binaural Impulse response Acquisition) wearing the acoustically transparent headphones to be deployed in the simulation was used for the recording. ITDs were extracted using the onset detection method explained in

45

5. Anthropometry-based ITD individualization

Figure 5.1.: Relevant anthropometric measures defining the individual ITD

Figure 5.2.: Listening test setup. Up: While using the reference speaker. Down: while using binaural system.

46

5. Anthropometry-based ITD individualization

Figure 5.3.: Low-pass filter applied to the noise-burst stimulus to minimize the lateralization influence of ILD.

sec.

2.1.3.2

, setting the detection threshold as -35 dB. A minimum phase BRIR dataset was

stored for its use in the simulation. A Genelec 1030A loudspeaker was positioned frontally at a distance of 2 m from HATS/listener.

The listening test consisted in adjusting the presented ITD by finding the point where the virtual sound source appears stable and matches the position of the real loudspeaker. To accomplish this task the listener could switch between real loudspeaker and simulation. All subjects were told to rotate the head in order to produce bigger ITD values that facilitate the calibration task. Figure

5.2

shows a schema of the setup.

5.1.1.1. Stimulus

Low-pass filtered white noise bursts were used as stimulus to minimize the lateralization influence of ILD which start to be a dominant localization cue at about 1.5 kHz. The filter characteristics chosen were:

Filter type: Butterworth

Attenuation: 40.0 dB

F pass

:

F stop

:

1.0 kHz

1.5 kHz

The magnitude response of the filter can be seen in fig.

5.3

.

47

5. Anthropometry-based ITD individualization

Figure 5.4.: Graphical user interface of the listening test audio application.

5.1.1.2. Listening-test’s software

To execute the test an audio application was written in C++. The software implemented following features:

• Triggering either loudspeaker or binaural simulation.

• Randomized starting value of the scaling factor between 0.0 (no ITD) and 2.0 (2x original ITD) at each trial.

• Storing decisions (calibrated scaling factors) of each trial in a computer readable format.

• Test status sent over the network to a monitoring remote computer.

• GUI.

Figure

5.4

shows the GUI of the application, while figure

5.5

show the protocol of the test for a selected user including randomized ITD start values and the generated scaling factors as comma separated values (CSV).

5.1.1.3. Interface

A computer keyboard with marked operational keys (see fig.

5.6

) was used as interface in

the listening test. The keys marked with plus and minus allowed the listener to scale the

ITD in -+1% steps of the original ITD. Although the ITD-I has no constraints in scaling resolution, this step size was chosen to keep the task of test simple.

48

5. Anthropometry-based ITD individualization

Figure 5.5.: Protocol of a selected user’s listening test written in CSV format. Red arrow points to the column of the generated ITD scaling factors

Listener’s could switch between real loudspeaker and simulation at any time. Once the appropriate scaling factor was found, the “apply" key allowed the user to store the value and start a new trial.

Figure 5.6.: User interface for the listening test

49

5. Anthropometry-based ITD individualization

5.1.2. Listening test procedure

Before the test began, the 4 head measurements mentioned in sec

5.1

were conducted on the 11 participants that took part in the test. Information about age, experience in listening tests, music studies and known hearing loss were gathered with following results:

Average age: 28 years.

Experience in: listening tests All.

Music studies: All but 1, 8 years average.

Hearing loss acc. to own testimony: None.

The participants were seated in the same place as FABIAN for the recording of the binaural dataset. Special care was taken on keeping the same height of the subject’s ear-canals.

All participants were instructed about the use of the interface and were introduced with the effects of non-individual ITD with the aid of examples (ie. exaggerating the ITD size to provoke instability of the virtual sound source at head movements). The listeners could practice the individualization task until they felt confident with the procedure.

Each subject had to scale 10 times its own perceptually correct ITD, always starting with randomized values. The test took 25 to 30 minutes.

5.2. Statistical analysis

Despite training and the obvious effect of the non-individual ITD, the calibration task was for some subjects difficult to realize. By means of residual analysis and outlier tests 2 of the

11 subjects were excluded from the final analysis.

Figure

5.7

shows the individual distributions from the 10 ITD scaling factors of each of the

9 subjects as box-plots. Note the big confidence intervals. The individual medians covered

−5% to +7%.

Table

5.1

presents a resume of the multiple regression analysis on different configurations.

50

5. Anthropometry-based ITD individualization

Figure 5.7.: Distribution of the individually generated ITD scaling factors from 9 subjects.

Note the big dispersions.

Explained variance (adj. R

2

) indicated that a single predictor the intertragus distance is best suited for predicting the individual ITD scaling factors reaching 66.0 % of explained variance.

Figure

5.8

shows that the individual ITD scaling factors increase linearly in proportion to the head diameter described by the intertragus distance. Figure

5.8

also shows all nine mean individual scaling values together with their 95% confidence intervals (CI).

The linear regression model is also shown with its 95% CIs.

5.3. Anthropometry-based ITD individualization formula

The regression formula for predicting the individual scaling factor based on the intertragus distance is:

S

= 0.00304 · d

i

+ 0.5792

(5.1) with the intertragus distance d

i

, specified in millimeters. Note that this model is valid only for binaural datasets acquired with the HATS FABIAN. The model could possibly be generalized to arbitrary binaural data by scaling the predicted values by a certain ratio of the

51

5. Anthropometry-based ITD individualization

Model

Intertragus distance

Intertragus distance,

Depth-DIN

Intertragus distance,

Depth-DIN, Height-

DIN

R R square

0.838

0.703

0.839

0.704

0.840

0.705

Adjusted R square

0.660

0.606

0.529

Std.

Error of the estimate

0.0155637

0.167640

0.183355

Table 5.1.: Summary of correlation factors using three models: Intertragus distance; Intertragus distance and Depth-DIN; Intertragus distance, Depth-DIN and Height-

DIN. Adjusted R square of the intertragus distance is bigger as in the other models.

intertragus distances of the foreign and FABIAN’s artificial heads. Though this aspect has not been investigated yet.

52

5. Anthropometry-based ITD individualization

Figure 5.8.: Modeling of listening test results: The linear regression model over the intertragus distance is shown with hyperbolic 95% CIs.

53

Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement

Table of contents

Languages