MaximumPC 2007 07

r & d BREAKING DOWN TECH —PRESENT AND FUTURE

62 MAXIMUMPC july 2007

W

indows Vista plays David Bowie

songs—and all other types of audio—

differently than Windows XP does. You won’t

see the differences, because the changes are

deep beneath the surface of the new OS, but

you’ll probably hear them. And that can be a

good thing—or a very bad thing.

Microsoft, you see, fi red most of the band

while designing Vista’s audio subsystem: The

company’s engineers rewrote and reorga-

nized the entire audio stack, moving nearly

all kernel functions to the user level. The ben-

efi ts of this move are twofold: First, any bug

that might creep into the audio system affects

only the audio system; it doesn’t bring down

the entire OS. Second, the change increases

the number of programmers who are qualifi ed

to develop new audio applications because

programming at the user level is much easier

than programming at the kernel level.

When you consider that the Windows

audio system has changed very little since

Windows 98, an overhaul was long over-

due. In the midst of this transition, however,

Microsoft dumped support for the custom

property set that enabled DirectSound and

DirectSound3D processing to be offl oaded

to a soundcard or even an integrated audio

system. The host CPU must now perform

these calls. (Microsoft says its research indi-

cates that only four percent of PC users have

soundcards, but we know that number is

much higher among Maximum PC readers.)

THE LONG AND WINDING ROAD

Over time, less and less software will rely

on the legacy Windows audio architec-

ture, and Microsoft claims that applications

properly designed to use DirectSound and

DirectSound3D should work fi ne in the new

environment. But without access to hardware

acceleration, many applications that utilize

widely adopted but proprietary advanced

audio APIs (application program interfaces),

such as Creative’s EAX, will revert to stereo

when running in Vista.

Microsoft worked with Creative Labs and

Nvidia for more than a year to develop a new

driver model that would enable hardware

acceleration in Vista but pulled out of the initia-

tive when Nvidia exited the audio-hardware

market. Creative then turned all its attention

to the development of the free OpenAL API.

Applications that use OpenAL behave the

same way in Vista as they do in XP: Calls to

OpenAL fall back to software mode (the code

is executed on the host CPU if no dedicated

soundcard is available). Creative is also devel-

oping a solution—called ALchemy—that will

translate DirectSound3D and EAX calls to

OpenAL in real time so that they can also be

executed in hardware.

Microsoft maintains that its changes

render Vista more stable than XP; what’s

more, it claims Vista’s audio system should

deliver higher fi delity than previous versions

of Windows. This latter claim stems from the

fact that the new audio system uses a 32-bit

fl oating-point pipeline, compared to XP’s

16-bit integer pipeline. If a DSP (digital signal

processing) algorithm causes a buffer over-

fl ow, the result is typically distortion and other

undesirable audible artifacts. This shouldn’t

happen with a 32-bit FP pipeline.

This doesn’t change the fact that Vista has

developed a bad reputation among gamers,

but it’s diffi cult to determine whether the oper-

ating system’s stability issues—not all of which

are associated with soundcards—are related

to the way games are coded for Windows

(i.e., poorly) or the way the OS itself is writ-

ten (ditto). Evaluating claims of audio fi delity

should be an easier task.

WILL IT GO AROUND

IN CIRCLES?

Vista’s audio architecture is based on three

major APIs: the Multimedia Device API (used

to catalog and manage audio endpoints,

such as microphones and speakers), the

Device Topology API (a tool programmers

use to interrogate audio hardware, such

as soundcards, so they can create a soft-

ware control panel that can access all that

device’s features), and the core of Vista’s

audio system, the Windows Audio Session

API (WASAPI), which actually renders and

captures audio streams.

Three smaller APIs take their place

beneath WASAPI in the Vista audio

stack: IAudioClient acts as a gateway

so that audio can stream between an

audio application and the Vista Audio

Engine. IAudioRenderClient is an interface

that audio-rendering applications use

to stream their output to an audio end-

point device. The last of these four APIs,

IAudioCaptureClient, is used by audio-

capture applications to record audio from

capture endpoints.

In what Microsoft describes as Shared

mode, each of these IAudio clients builds

a stream in the audio engine, and then

all these streams are mixed in the mix

buffer. The mixed audio is then passed

to the kernel-streaming endpoint. This is

White Paper: Audio in Vista

HOW IT WORKS Windows Vista Audio Stack

Microsoft had two key objectives for Vista’s audio system: Move as much of the process out of Kernel mode as possible and reduce

latency by signifi cantly streamlining the entire audio stack.

Cha- cha- cha- changes:

Microsoft re-architects

Windows’s sound system

BY ZACK STERN WASAPI

IAudioRenderClient

IAudioCaptureClient

IAudioClient

AUDIO ENGINE

MIX

BUFFER

User mode Kernel mode

Audio Stream

Audio Stream

Audio Stream Wave RTMini Port

Vendor

Miniport

Driver

DMA

Memory

Soundcard

Speaker

Soundcard