Last update: 2001/3/9
  8 Explanation on the database


Measurement condition

The measurement was carried out in the anechoic room of the Institute of Technology, Shimizu Corporation. Setup of the measurement system is illustrated in Figure 1. The subject sat on the floor made of glasswool sheets with his head fixed to the head rest. Sound signal was radiated from the loudspeaker attached to the traverser fixed to the ceiling of the anechoic room. Arm of the traverser can be rotated in arbitrary angles, and the position of the loudspeaker can be moved in arbitrary position along the arm. This enables the loudspeaker to be positioned in arbitrary position on the upper hemisphere around the subject. The radius of the sphere is 1.2 m. Time Stretched Pulse (TSP) was used as the sound signal radiated from the loudspeaker. Small electret microphones were adjusted to the entrances of the subject's both ears using thin metal wires. Output of the microphones was amplified by the hand-made amplifier, then A/D-converted to the digital signal processor unit, where the inverse filter of TSP was convolved with the response. Sampling frequency was 44.1 kHz, and the measured responses were time-averaged by eight times per a single measurement point.

Measurement system
Figure 1: Measurement system
(Click of the figure shows the larger one (JPEG, 427 kB))
EPS+gzip file (5849 Bytes)
Source position
Figure 2: Measured source direction (Top view)
(Click of the figure shows the larger one (JPEG, 206 kB))
EPS+gzip file (2609 Bytes)

Contents of the measurement

In the conditions mentioned above, HRTFs of three male subjects in almost whole directions on upper hemisphere were measured. Positions of the sound source were depicted in \reffig{source-position}. As shown in this figure, interval of azimuthal angle was changed depending on the elevation, which was set at every 10 degrees. Change of interval angle with elevation is summarized in \reftbl{interval}. Resulting number of measurement points was 430. Since the entire measurement needed too long time for each subject to sit still, eight to nine rests were inserted during the measurement. As a result, about four hours were spent per one subject for the measurement of the whole set of HRTFs.

Table 1: Change in number of measured angles with elevation
Elevation [deg]Angular step [deg]Number of points
0572
10572
20572
30660
40660
50845
601036
701524
803012
90---1
Total454

About time windowing (added on 01/05/29/)

The rectangular window of 512 points long was applied to each measured response. The time windowing was so positioned that the sample with the maximum amplitude came around the center of the window while keeping the relative timing/delay among the impulse responses intact. More detailed explanation of the windowing is as follows:
  1. First, the impulse response with the longest delay is selected. This usually corresponds to the direction of opposite side of the ear measured. Then, the center of the window was set to the position of the sample with the maximum amplitude.
  2. Then, the same positioning was applied to the other responses by keeping relative time between responses. This enables to keep the relative timing/delay among the impulse responses intact.

References

  1. J. Blauert, Spatial Hearing (MIT Press)
    ...Definition of HRTF, spatial hearing
  2. Yôiti Suzuki, Futoshi Asano, Hack-Yoon Kim and Toshio Sone, "An optimum computer-generated pulse signal suitable for the measurement of very long impulse responses," J. Acoust. Soc. Am., 97(2), 1119-1123(1995).
    ...on Time-Stretched Pulse (TSP)