New 'uncompressor' design will be coming soon.

2018-06-08 03:08:04

I have been working on extracting some of the technology from the aborted DBX Type 1 mess, and bolted the detector onto the front of the 'uncompressor' (used to be called the restoration processor by me.)  So, instead of doing a pure DBX decode after the DolbyA compatible decode, I am now doing the expansion (restoration) of the dynamic range using the uncompressor.

The 'uncompressor' is not you grandparents DBX 5BX at all -- far, far beyond that.   The input detector is now close to a DBX/THAT style RMS input instead of my own purely calculated RMS -- and I have to say that the RMS/LOG style of detector blows away my detector -- so I have semi-permanently adopted it.  I think that it will be in the final version -- coming up very soon.  This compression removal tool (not just an expander) has been promised in short weeks if not days.

I have a repository of examples, and a short 30second snippet posted here that shows how clean the results can be.  I tried to look around to find a good example where the results of compression removal can be clearly descerned in 30 seconds, and couldn't find any other than this incomplete example that is attached.

I also have a pointer to a repository that is expected to live only a few days -- but really shows the impact (I don't have any before examples online -- the results are clear enough -- especially listen to the entire Nat King Cole example.)

The source material for most of the examples was the ABBA studio TCSR CDs.  If you want to hear the 'before' then listen to those CDs -- those were EXACTLY what I used.  Typical crest factor on that CD is in the range of 4-6 at best.   The crest factor of most of these results is at least 6, and mostly in the range of 9-10.

When comparing with the orignial vinyl, these results have more presence and clarity (much less rumble also), but the dynamics are slightly restricted.  When compared with the ABBA discography version (also available for sale), the crest factor is also better -- but the discography wasn't quite so damaged.  I would tend to purchase the discography again if I couldn't clean up the ABBA TCSR -- which becomes vastly superior once cleaned up.  BOTH THE discography and TCSR verisons that I got are definitely DolbyA encoded.

IN all cases, except the vinyl -- the originals of both the TCSR and discography were DolbyA encoded with APPARENTLY approx 3dB cut at 3kHz and 3dB cut at 9kHz.

These results will demonstrate how good the 'uncompressor' can be when used in conjunction with the DolbyA compatible decoder.

The short ABBA example is attached, and several more complete examples are online for a few days (and are actually enjoyable listening.)

The 'uncompressor' is coming out FOR FREE very soon (like I wrote above short weeks if not days.)

The demos could stand a little bit of EQ -- I wasn't concentrating on producing a final version of anything -- so you might have to EQ to taste.  (I'd probably soften the high end a little, and bring up a few dB at 240Hz and below.)

Demos:  https://spaces.hightail.com/space/b76V0730pL

John

Superior demo versions -- the mp3 encoding didn't do demos justice!!!

Reply #1 – 2018-06-08 07:23:17

I just listened again after not wearing headphones for a few hours -- and I do suggest dropping the high frequencies between 10 and 20kHz about 3-4dB -- perhaps -1.5dB at 10kHz and -4.5 at 20kHz, and a smooth decrease between.   The highs are a little metallic on the demos -- probably partially due to the mp3 encoding -- there is more than enough HF in the originals -- the original FLAC versions are a little more silky smooth with clearer 's' sounds.   On the low end it seems like shelving up about 2-3dB at 240 sounds pretty good -- maybe starting at 350 instead of 240 might be a little better.   The lowest lows are adequate -- absolutely no rolloff, so the 50Hz region is adequate, especially once the lows are shelved up a bit.  (Hightail doesn't directly support the more extreme mp3 encoding for direct listening -- so that is why I am trying to push so much intense HF content through 200kbps.  Normally, I'd try to use 320 mp3 or at least 256opus.)

In order to communicate how good the original .flac versions sounded instead of these .mp3 demo versions (MUCH SMOOTHER)...  I uploaded two flac versions -- much bigger, but very, very pretty!!!!
Somehow my .mp3 encoding lost A LOT in the translation....

https://spaces.hightail.com/space/UntM4LCdcm

https://spaces.hightail.com/space/wU6nwJD4bW

This final one won't play directly for some reason -- but you can even tell that SOS is cleaner sounding than the mp3 version:

https://spaces.hightail.com/space/ko2yTjF5YY

New uncompressor - better demos

Reply #2 – 2018-06-08 11:52:25

I just uploaded cleaner versions of the demos.  I was pushing mp3 a little too hard, causing the results to be too metallic sounding.  I did a bit of rolloff producing some new demos, but didn't touch anything below about 16kHz -- and the results were much cleaner.  I didn't know that mp3 would be so unpleasant sounding when maintaining a full signal level with no rolloff.  Might also have been a bit of Gibbs ringing -- didn't think about that as an issue.  I have been very careful to avoid aliasing & intermod in the various software projects that I have been writing, but didn't consider that sometimes other software might not be able to deal with the intensity.   I was starting to play with opus -- trying to see if I could find a combo where Hightail would allow for direct play, but couldn't find an opus version that would work with the direct play.  SO -- I reviewed how I was encoding the mp3, and simply left a softer rolloff -- the same intense flat to the highest frequency didn't cause trouble with flac encoding -- must be something to do with the lossy choices with the lame mp3 encoder...  Oh well -- these are much prettier.   I didn't attach a short demo -- the value is so limited because the processing is difficult to discern with short demos.   Again, these demos will disappear within a few days -- right now I am working on a releaseable version of the DolbyA decoder with better 192k support along with a releasable version of the uncompressor.

Anyway, these demos are much nicer (smoother, less metallic) sounding now -- same location -- encoded from EXACTLY the same .flac files as before:

https://spaces.hightail.com/space/b76V0730pL

Re: New 'uncompressor' design will be coming soon.

Reply #3 – 2018-06-08 14:30:56

Attached is a list of dynamics measurements from sox  -- showing numerically the improvement using the uncompressor.  In all cases, the uncompressor has an 'easier' sound -- less compression than the worst, but cleaner sound than the vinyl.

In almost all cases, the vinyl has the widest dynamic range, but the uncompressor comes pretty close when compared with the extreme of the more recent ABBAstudio and ABBAdiscography.

The real advantage of the uncompressed version is that the sound is generally pristine, but not compressed as badly as most common unprocessed distributions nowadays.

A crest factor below 5 is too compressed (IMO) for almost anything, 6 or higher for a crest factor tends to be more natural for rock/pop.  Also a peak to RMS less than about 15dB starts getting to be too extreme -- I prefer at least 16dB if possible.  In some cases, the 'uncompressor' didn't 'uncompress' as much as it could -- but I used the same settings for every song.  In many cases, an addition crest factor increment or dB or so of peak-to-RMS would be achievable using the 'uncompressor' with further sound quality improvment.

I simply used a conservative setting on the uncompressor for this test run.

John

Re: New 'uncompressor' design will be coming soon.

Reply #4 – 2018-08-21 03:19:12

I've been reading about your Expander (or Uncompressor). Looks like you have put a lot of work into it.

My interest is - some years ago I ripped all my mag tapes and cassettes while my hardware was still working. As you would appreciate the dynamic range of these tracks is limited - perhaps your method could bring new life...

Just for a bit of understanding - does your method split the source into multiple bands, calculates an exponent for each, then combine them (somehow) into a single exponent value?

I haven't downloaded your software. The computer that I use for media (audio) is a Win7 32bit. I do have a 64bit Linux (Suse) machine, but it would be a real pain to set it up for this purpose.

Is there a particular reason why the process runs slow - language, compiler, interfaces?

Stacker