Compression tricks of the trade

Ok, fixed my audio issue... and now to a new problem. I have a problem with choosing compressions, every time I've tried to use a codec to make my file size smaller than gigantic, I exhibit a HUGE loss of quality. What are your tricks to getting the best out of your codecs?

 

It is far from extensive, but check out the page on compression on the TFN main page, it may help to begin with.
Click here if the walk is to far

 

I recently squashed a 4 minute video to about 9MB. I used Adobe Premiere's "Get Properties For > File" function under the File menu and exported this info to a text file, to keep a record of good settings I can refer to in the future:

File size: 9.01MB bytes
Total duration: 0;03;53;14
Average data rate: 39.59KB per second
Image size: 180 x 121
Pixel depth: 24 bits
Frame rate: 15.00 fps
Audio: 24000 Hz - 16 bit - Stereo

QuickTime details:
Movie contains 1 video track(s), 1 audio track(s) and 0 timecode track(s).

Video:
There are 3499 frames with a duration of 1/15th.

Video track 1:
Duration is 0;03;53;14
Average frame rate is 15.00 fps

Video track 1 contains 1 type(s) of video data:

Video data block #1:
Frame Size = 180 x 121
Compressor = Sorenson Video 3
Quality = Normal (3.00)
Temporal = Most (5.00)

Audio:
Audio track 1 contains 1 type(s) of audio data:

Audio data block #1:
Format = 16 bit - Stereo
Rate = 24000.0000 Hz
Compressor = IMA 4:1

 

Yikes. 180 x 121 at 15 FPS..... No wonder your filesize is good.

But yeah, I too always use QuickTime Sorenson Video 3 (at medium or high) and IMA 4:1 (usually mono) for sound, and am often praised for being some sort of compression wizard. I'm not. QuickTime is free, not hard to use and produces very good results in quality and filesize.

Edit: Forgot to mention I export from AE using the render queue. Something I noticed was that when I export the same thing with allegedly the same compression out of Premiere, the filesize is noticeably larger the image quality is not quite as good.

Edit Again: Another thing I remembered I probably should mention since this is a tricks-of-the-trade thread, is that Sorenson has the peculiar tendency to darken the image a bit when it compresses. This is really the only weird little problem I've ever encountered using Sorenson, but it's remedied easily enough by simply bumping up the brightness of your shot(s) a few percent before you export for the web.

 

So in other words, it's compressing luminance data considerably. I noticed this with my logo sequence... the 1s and 0s in my "avatar digital" logo shot are very dark... and the areas where they were a little darker, are almost completely black.

I'm still playing around with different settings. Keep looking at my multimedia link I posted earlier for updates. If you keep your mouse pointer/cursor on top of the text under each clip, a small javascript popup will appear (at least on IE 5.0 and later) with more info on the clip's settings.

Granted the new chroma key test is a lot less data than the small video because it's much shorter, and there's very little actual motion... but I also did audio-only tests to see if I could optimize both the perceived output and the bandwidth of the audio by using a lowpass filter to cut out frequencies above 5kHz on a voice track. The MP3s appeared to be no different, but the original AIF files were considerably different in file size if I recall.. I will have to check on that, though.

Some of this I picked up by working in DSL/telephony. Telephone transmissions have a 700Hz high-pass and 3kHz low-pass filter to constrain the frequency range and digital bandwidth they occupy over fiber optic networks.