> Different schools of thought, for sure. After reading Frank
> and Bob's writings on the matter, my impression is that with
> heavily coded and data-compressed audio, the name of the
> game is playing the CODECs to produce a quality coded result
> rather than worrying about re-eq'ing and recharacterizing
> the spectral balance of the source material. To that end, it
> seems to me that you start out with clean source material
> and lightly compress/expand/or limit to get the levels
> consistent and protect the encoder, and if you do that, you
> should have outstanding results. If time and attention is
> paid to the quality of sources, then the audio should
> already be high quality audio and you shouldn't need to do
> multi-band processing. From what Bob and Frank have written,
> the more density and processing you put in the audio before
> the codec, the more likely you are going to get garbage.
> Most major-label recordings are already processed to an inch
> of their lives. In our time streaming, we've discovered that
> even adding a relay in our shoutcast stream will start to
> degrade the audio. It probably shouldn't be so, but it is.
Yes, you have to consider the codec when tweaking the processing. This is particularly true for high frequencies, where the codecs usually exhibit much of their artifacts. But if you're listening to your stream when tweaking (rather than the direct output from the processor) you are automatically taking codec into account. And the whole transmission, as a matter of fact. To some small extent, this is similar as different exciters having different sound on FM radio.
And yes, the quality of sources is extremely important. Just like with FM radio, but even more so because you're dealing with significant perceptual reduction at the end. You definitely want to have linear, CD-quality sources and you definitely want to avoid coded audio. Overall, the cleaner your sources are, the better you will sound. Noticeably.
But as far as processing is concerned, the amount of compression is not critical. You can have similar compression density as with FM radio. As a matter of fact, most codecs sound better when constantly fed with high-level audio (which compression provides). When the levels drop, codecs start to sound phasey and flanging on high end. When there's constant high-end present, it masks the artifacts and audio sounds better. This is particularly true with older generation streaming codecs, like Windows Media and Real Media. So by tailoring multi-band processing to complement codec, you can actually have better audio then you would if you were using wide-band processing. Or no processing at all.
Now, what Frank and Bob are actually telling you, is that you don't want to clip your audio! Clipping produces harmonics. These harmonics not only distort audio, but also contaminate spectrum which codec "wastes bits" on, that could be used to code more of relevant audio information. Clipped audio sounds much worse when coded, than clean audio does. This is why all processor for digital delivery use look-ahead limiting for peak control. Look-ahead limiting produces no extra harmonics, only IMD (if you lean on in too hard).
> We're running a direct feed from our analog DA into a
> compressor/expander/limiter and then into the hardware
> encoder, and it sounds tight, clean, and consistent...
>
www.wxryfm.org. I can't see why we would ever need to do any
> further processing for the stream because it would be
> overkill.
That's your position and I respect that. Others might feel they want to do more with their webstreams. Whether there will be a noticeable difference, as always, will be judged by listeners ;-)
Regards,
Goran Tomas
<P ID="signature">______________