Description of HRTF data set Fred Wightman and Doris Kistler University of Wisconsin-Madison Thank you for your interest in our data. We are delighted that they can be of some use. By now you should have returned the University of Wisconsin Material Transfer Agreement forms (if you haven't done so, stop here), so you are free to take the data and use them in your research projects. There are HRTF data from left and right ears of 5 subjects in this directory. There are 505 time-domain impulse responses in each of the files. Each impulse response is represented by 256 time points, at a 50 kHz sample rate. (The HRTFs were originally measured at 100 kHz, and downsampled by simple decimation after processing and low- pass filtering.) The impulse-responses are stored one-after-another in the files, with no break between one 256-point impulse response and the next. All of the data are in floating-point form. The ASCII versions of the data, in E14.7 format with one time point per line, are in the files labelled XXX_Y.Txt, where XXX represents the subject and Y represents the ear (L or R). Since these files are large (more than 2 MB each), we also offer compressed ASCII versions (compressed with the shareware package PKZIP), in files labelled XXX_Ytxt.Zip. The compressed ASCII versions are a little bigger than 700 kB each. You should be able to use PKUNZIP on any computing platform to decompress the files. Finally, since you may eventually use the data on a PC (Intel), we offer the data in compressed binary form (single-precision PC floating-point format). These files, labelled XXX_Ybin.Exe, are about 475 kB each, and are "self-extracting". On any PC, simply run the *.Exe file and the uncompressed file will appear, with a name like XXX_Y.Bin. In order to transmit the files properly, you must set the file type to ASCII before you FTP the *.Txt files, and to IMAGE or BINARY mode before you FTP the *.Zip or *.Exe files. The file called HRTF.Map is a guide to what is contained in all the other files. It is an ASCII file that describes the azimuth and elevation coordinates (distance was constant at 1.4M) at which the HRTF measurements were made for each of the 505 records in each of the other files. The HRTF.Map file has 3 columns, the first gives the azimuth (degrees), the second gives the elevation and the third gives the sequence number in the file. In our coordinate system, zero azimuth is straight ahead, positive azimuths to the right, and negative azimuths to the left. Zero elevation is on the horizontal plane through the subject's ears, positive is above the plane, and negative is below the plane. This is a "single-pole" coordinate system. Measurements were made at 10 degree intervals for both azimuth (ranging from +180 to -170) and elevation (ranging from +80 to -50). The single measurement made at 90 degrees elevation is first in each file. The impulse-responses represent "raw" HRTFs, measured with a probe microphone close to the subject's eardrum. Other than the usual processing to remove the characteristics of measuring stimulus, loudspeaker, and microphone, there has been no further processing of the data. Because of standing waves in the ear canal, which produce a notch in the magnitude response at a frequency where the probe-eardrum distance is 1/4 wavelength, you should be cautious about interpreting the data at frequencies above about 15 kHz. If you wish to use the data to synthesize virtual sound sources for presentation over headphones, you should divide (in the frequency domain) the magnitude component of each response by the "diffuse- field" response, which is the RMS average of the magnitudes at all 505 positions. This produces what is commonly called a "directional transfer function" or DTF. Since for all practical purposes the HRTFs are minimum-phase, it is both convenient and approximately correct to compute the transfer functions of your FIR synthesis filters by coupling the magnitude of each DTF with a minimum-phase phase response. Depending on the quality of headphones used, you may need further processing to correct for headphone response. If you are using headphones that claim to be "diffuse-field equalized", further processing may be unnecessary. In case you are interested, here are the heights and genders of the subjects from whom we took the HRTF data: AFW F 68.5" SJX F 68.0" SOU F 65.0" SOS M 74.0" SOW M 75.0"