Got a really obvious comparison between different sources of the same song with the results of the processing software that I have been working on. Note that I am using ABBA here, but this isn't just ABBA related, but other processing examples can be quite obvious in their differences. ABBA is just so darned problematical to process and get good results -- so it is one of my worst case test subjects (even though I do like the music!!!) This example is interesting because I have several published versions, plus two versions that the 'so called' pseudo-DolbyA & 'restoration' packages produced. I STILL NEED A NAME FOR THE 'restoration' processor. Any help is happily accepted!!!
Each example is limited to 30seconds.
1) original version of recording that was used for processing source - tcom-orig.mp3
2) processed by 'pseudo-DolbyA/restoration' processor -- tcom-jdraw.mp3
3) processed like (2) above "tcom-jdraw.mp3" was, but additional compression/limiting step -- tcom-jdcomp.mp3
4) normal 'American' CD release of ABBA Gold -- tcom-CDusa.mp3
5) Polar European release (usually deemed one of the best ABBA) -- tcom-polar.mp3
I tuned the versions produced to sound similar to the Polar release (not perfect, of course.)
The 'restoration' processor was set using somewhat default settings (psuedo-DolbyA, 1:1.33 expansion, maximum transient recovery (takes lots of CPU), and running only at 48K because of time constraints.) The psuedo-DolbyA is available in source and binary right now, and the restoration processor is available upon request within a few weeks -- might start distributing it within weeks also. (I keep re-optimizing the 'restoration' processor, and haven't even created a proper name for it yet!!!) The 'restoration' processor adapts very well to different kinds of music, and has only a few knobs (expansion ratio, threshold, and maybe one or two other things.)
For the compressor/limiter, only two phases of 8 band compression (both RMS, with differing gain range & attack/decay character) -- both are totally dynamic attack/decay, both with only 1.25:1 compression, and then a prelimiter (single band 4:1 compression, fairly fast, RMS-like), and the a true limiter complex. The prelimiter is hit by a few (1-3) dB much of the time, the true limiter is hit every few seconds. The compressors are running at approx 4-8dB of AGC, over a wide input range due to the very low compression ratio. (The compressors are fairly conservative sounding up to 2:1 compression with 5-10dB of depth each, the prelimiter/limiter complex fools me into thinking that it isn't working because it has so few artifacts when properly used.) All of the compressors use special anti-intermodulation methods. The compressor/limiter is in the midst of a rework, and probably won't be available for approx 6months.
(Sorry about the abrupt music termination.)
My own observations is that the commercial versions were compressed so much that they sounded almost 'nasal' when compared with the 'expanded' versions. I am still sometimes tweaking the 'big' processor, and I have one major programming step -- separate out the 'expansion' code and more into a seperate program. I have that 75% done, but havent' completed it yet.
Have one more example -- this one has been slightly rematrixed to make the voices a little more present. I know that this requires much more work -- it would be a few hours to get it perfect, but if you compare this with tcom-jdraw.mp3, you'll hear a major difference in presence.
These examples show how much positive that can be done with the right software. The jdraw result is definitely NOT raw, but a simple raw output of the processor.
The improvement from the Polar release to the jdraw or jdcomp versions is VERY VERY substantial. If you look at the USA CD version, it is attrocious compared to even the Polar release.
I am demoing something that is NOT any better than what the original ABBA people created, but all too often, the disk mastering destroys or distorts the original work.
My results are NOT perfection, and I'd probably spend an hour or two on each song at least if I was doing this commercially (to get it absolutely correct.)
Again, the improvement from Polar to jdraw or jdcomp is very, very substantial. Some people might like this version better (or realize that a few hours of tuning could result in something VERY nice.)
The voices in this version are MUCH MUCH clearer -- and just a little tuning/positioning and equalization might be fruitful.
Got a somewhat good reception on the ABBA chat group on these examples. If you listen to the files whose names end in 'Clean', there is some very good processing going on -- very clear sound. The Dreamworld example is amazing -- apparently it got compressed several times in succession, practically destroyed. The result is quite listenable, but does show damage.
I would be doing more 'scientific' comparisons, but the differences & improvements (& problems) are so extremely obvious, that fine grained audio measurements aren't really needed. Also, the gross level noise reduction can be seen with a spectrogram, even though the exact amount of NR isn't really measurable there.
These examples are disappearing in a few days, but the audio processor (the 'restoration processor') is now frozen, and I am repackaging the software moving the software into its own binary. The psuedo-DolbyA has been frozen for a few weeks. Binaries will be available in a week or so, and source code in a couple of weeks after that. The VERY GOOD news is that the source code/function is now frozen!!!
Here are where the examples are -- again, the filenames with the 'clarification' have 'Clean' in them. Some normally processed stuff is in there also. Nothing in there is 'boring.' There is a Carpenters example that is super clean also: