The problem of hiss on the audio of recordings made with ATI video
cards and C-Media audio implementations has been floating around
various groups for a while now.

A few factors conspire to produce this problem.

1) The audio ouput of the offending ATI card contains excessive noise
that is outside of the audio band. Measurements of an all-in-wonder
7500 show noise extending to hundreds of kilohertz with spectral
density that rises to be many dB higher than the level of the in-band
audio noise. (This is a normal consequence of some types of audio
processing and circuitry that does this is said to be doing
"noise-shaping".)
2) The offending C-Media 8738 audio chip does not employ an
oversampling type of analog-to-digital converter.
3) The manufacturer (Soyo, Asus, etc?) that uses the C-Media chip
connects the audio into the chip without any input filtering.

Under these conditions, the C-Media a-d converter creates a
substantial audio hiss because it aliases a huge bandwidth of
huge-amplitude out-of-band noise down in to the audio band.

Any digitized audio will have the hiss. It will show up on recordings.
It will also show up when monitoring the "record" half of the C-Media
chip - I see it easily listening to Roxio Spin Doctor, and can see it
clearly with the TrueRTA audio analyzer (by True Audio).

Audio that passes through the system without being digitized will not
have the excess hiss. This is what happens when the ATI tv tuner is
fed to the audio line-in and listened to 'live' through the line-out
-- it sounds just fine. The C-Media pass-through path is very wide
bandwith, but your ears will not respond to the high-frequency noise
even if your speakers could reproduce it.

On the Soyo SY-K7V Dragon Plus, the hiss is not noticeable when (one
channel  of ) the ATI audio is fed to the mic input. Measurements show
that the mic input is filtered, and its frequency response begins
falling off at 6dB/octave a little above 10kHz. That is the right
thing to do. In contrast, the line input shows no indication of
filtering - put in a 45.1kHz sine wave at 44.1kHz sampling rate and
you will get a perfect digital 1kHz tone just as though the sine wave
was 1kHz (45.1-44.1=1kHz). Same goes for a 89.2 kHz sine wave
(89.2-2*44.1=1kHz). And on it goes.

Whose fault?
- ATI should have included filters to knock down the out-of-band noise
that its chips produce.
- Soyo should have provided anti-aliasing filters at the input of the
C-Media chip. The ATI situation is particularly pronounced, but all
sources have some wideband noise, and lack of filtering means that a
recording ends up noisier than it has to be because of the aliasing.
(Either one would do the trick, and one manufacturer could probably
blame the other, but both could be said to be deficient.)

How to fix?
- For sources like vhs tape, don't run the audio through the ATI
dongle. Unplug the ATI audio and run the vhs audio straight into the
line-in. If the vhs is clean, this will make the noise about 20dB
better, and the signal level will actually be higher (audio through
the ATI card loses about 10dB of signal), for a 30dB improvement.
Can't do this for the tuner, though.
- Add your own filtering. A .005uF capacitor across each (L and R) of
the ATI outputs will make a noticeable difference, but it won't
eliminate the problem.
- Run the ATI analog audio through an good external a-d device and use
the C-Media s/pdif input.
- Switch to an audio card that takes care of out-of-band stuff (proper
anti-aliasing, better when combined with an oversampling a-d
converter).

What's not going to help (much): 
- Drivers - they're not the problem, unless they could reach out and
do something about analog (continuous-time) hardware filtering.
- Fiddling with audio levels.
- Fiddling with the recording itself. Once the noise is sampled into
the audio, you can't affect it without affecting the signal as well.
You have to catch it while it's out-of-band before it's aliased.

And just to be definite - this is not a full-duplex or 8-bit audio
problem. Examine the contents of a recorded .wav file - it shows the
samples filling all 16 bits.

As a final note, the Soyo implementation of the C-Media analog
line-out has no filtering either, so a digital source (.wav file,
s/pdif input etc.) gets reproduced as a staircase waveform (no
reconstruction filters). This is not much of a problem if all you do
is listen to it, or even if it goes into a digital system that limits
itself to the audio band.

(If you want to see some of this, get the free copy of TrueRTA and
monitor the spectrum of the line input with and without your ATI card
connected.)  www.trueaudio.com

The above observations were made on a Soyo SY-K7V Dragon Plus with an
ATI All-in-wonder Radeon 7500. The explanation should apply to 
systems with similar hardware that have similar symptoms.

Hope this helps.

Ed.