r&dBREAKING DOWN TECH —PRESENT AND FUTURE
66 MA XIMUMPC JANUARY 2006
Y
ou’ve no doubt heard or read the
term “head-related transfer functions”
(HRTFs) many times. Indeed, soundcard,
speaker, and headphone manufacturers
bandy the acronym about as though its defi -
nition were common knowledge. So what
exactly are HRTFs, and what benefi ts are
they capable of delivering?
HRTFs are best thought of as fi lters
that describe what happens to sounds as
they arrive at your ears from various direc-
tions. Once you have these fi lters available
for audio signal processing, they provide a
means to position virtual sound sources at
arbitrary positions in the space around your
head by tweaking the signal (hence the term
“head-related”).
This technology has had a substantial
impact on PC-audio applications in the last
10 years, especially in relation to interactive
games and multichannel surround-sound
playback via headphones and stereo loud-
speakers. This white paper explains how
HRTFs are used to create such illusions.
It will also examine the technical details of
HRTF-based processing in two common PC
sound applications: Interactive positioning
of sound for games, and the mixing of 5.1-
channel sound into just two audio channels
that nonetheless allow you to distinguish the
spatial direction of the sources as if they were
arriving from fi ve surrounding loudspeakers.
Just to make sure that readers are all
on the same page, we’ll begin with a brief
description of the foundations of HRTF tech-
nology in acoustical science and digital sig-
nal processing (DSP), before proceeding to
discuss its application to practical problems.THE ORIGIN OF HRTFS
HRTFs come from acoustical measurements
designed to capture how the sound that
reaches your ears changes as the direction
from which the sound arrives changes. The
simplest way to measure this is to record
(using a tiny microphone inside your ear),
the response to a brief impulse (e.g., a click)
produced by a loudspeaker. When you com-
pare the recordings between ears, you dis-
cover that the head blocks high-frequency
sound pretty well, as if casting an acoustic
shadow on the side of the head facing away
from the speaker. Including effects of the
outer ear, these head-related fi ltering effects
cause a change in the amplitude of sound
reaching your ears at different frequencies,
allowing you to easily determine the loud-
speaker’s position in 3D space.
How can digital fi lters be created to
capture these details, which are thought
to enable you to localize sounds in the
world around you? The fi rst step is to take
your head out of the picture and record
the impulse produced by a loudspeakerusing the tiny microphone that you later will
put in your ear. When you put your head
back in the picture, we can see what has
happened to the impulse that the micro-
phone picks up.
As the frequency of the sound rises
above approximately 2kHz, the sound
amplitude (essentially the volume) will start
to get higher at the ear facing the speaker.
At the same time, the sound at the other ear
will start to get much lower in amplitude.
But because the head doesn’t block low
frequencies so well, the sound from a sub-
woofer placed beside you will sound almost
exactly the same at both ears, although the
sound will arrive at the far ear a little later
than it does at the near ear. To give a better
overview of the nature of these changes, the
illustration above shows examples of HRTFs
measured at a listener’s right ear for fi ve
loudspeaker locations. The left-hand panel
displays changes in the shape of a perfect
impulse due to the acoustics of the head
and outer ear. The panel on the right shows
changes in frequency response. The upshotWhite Paper: Virtual 3D Audio
Audio engineers use HRTFs
to mix fi ve or more audio
channels down into two while
maintaining the spatial imag-
ery of multichannel surround
sound. Here’s how they do it
BY WILLIAM MATENSHOW IT WORKS Head-related transfer function in action
MicrophoneTIME (ms)icrophoneÑThe center panel in this illustration depicts a listener’s head with a microphone positioned in the right ear. Five
loudspeakers are placed around the listener. The panel on the left shows impulse responses, defi ned in terms of what
happens at the right ear in comparison to a perfect impulse (which would look like a single vertical line at time zero).
The purple curve on the left shows the impulse response for the center-channel speaker, with a direction identifi ed as
the 0-degree azimuth angle. The blue curves correspond to speaker directions 30 degrees removed from the frontal
direction, and the red curves correspond to the speakers removed by 110 degrees.
ÑAlthough each of the speakers is equidistant from the center of the listener’s head, the impulse arrives earlier from
the right-rear loudspeaker (the top red curve, displaying the response for +110 degrees). In fact, the impulse produced
by the left-rear loudspeaker (bottom red curve, displaying the response for -110 degrees) arrives almost a full mil-
lisecond later. In the right-hand panel, the corresponding frequency-response curves show that the amplitude of the sound
reaching the right ear is quite a bit greater between 5- and 10kHz when it arrives from the right-rear loudspeaker, than from
the left-rear loudspeaker. (Note that these frequency-response curves show changes in amplitude relative to the mean HRTF
amplitude curve, for which HRTFs were averaged over many directions to get a single amplitude curve over frequency.)-.5 0 .5 1IMPULSE RESPONSES SURROUND SOUND SPEAKERS FREQUENCY RESPONSES+110 ̊
+30 ̊
0 ̊
-30 ̊
-110 ̊+110 ̊
+30 ̊
0 ̊
-30 ̊
-110 ̊FREQUENCY (KHz)1 2 5 10