Generic Measured Datasets

As we mentioned in the previous section, measured HRTFs from one individual can be relatively effective in generating spatialized sound for another individual. This method is not as effective as using an individual's own HRTF measurements, but it may be a practical solution for many applications. It would increase the effectiveness of this approach if the physical characteristics of the listener were similar to those of the subject whose HRTFs were measured. One possible solution would be to have a number of HRTF datasets available from individuals whose characteristics represented a range of those found in the general population. In this way, a listener could be fit with the HRTF dataset that most matched their own physical characteristics. This is analogous to picking shoes from a limited selection of sizes.

KEMAR (Knowles Electronics Manikin for Acoustic Research) is an adjustable mannequin designed for making HRTF measurements. The research group at MIT Media Lab have made extensive measurements using KEMAR, and some of those datasets are publicly available for research use.

The measurements were made by placing microphones inside the mannequin's ear canal. Then broadband acoustic stimuli were presented from specific locations in free space. From this information the HRTF for each measured point is easily calculated. This dataset consists of 710 measurements collected over a wide range of spatial locations.

Our MATLAB script calls the code provided by the MIT research group to return the appropriate HRTF for a specified location. We then generated a stimulus and applied the HRTF data to the stimulus. The FFT was used to more efficiently compute this convolution.

Here is the script.

The script generates a wave file with sampling rate of 44.1kHz.

Back