Can Bad Recordings be Fixed?

[fas42]

Hmmm, I slowed down a bit - had a heavy session with the dentist yesterday, and wasn't feeling brilliant for a while ...

 

Anyway, to get a feeling for what before and after waveforms can look like, here's sandyk's original version, attenuated by 7dB:

 

 

and the restored waveform, latest effort:

 

 

Pretty close to optimum, but I still want to play around with it a bit more; there's a slight volume pumping I want to eliminate.

[Audiophile Neuroscience]

1 hour ago, fas42 said:

Hmmm, I slowed down a bit - had a heavy session with the dentist yesterday, and wasn't feeling brilliant for a while ...

 

Anyway, to get a feeling for what before and after waveforms can look like, here's sandyk's original version, attenuated by 7dB:

 

 

and the restored waveform, latest effort:

 

 

Pretty close to optimum, but I still want to play around with it a bit more; there's a slight volume pumping I want to eliminate.

 

Look forward to hearing it.

[tmtomh]

I believe you absolutely can effectively decompress, or partially decompress, a squashed and brickwalled mastering - but only in some cases, and it's often hard to predict ahead of time. Too often you get the result @sandyk posted a screenshot of earlier in the thread - same basic buzzcut brickwall waveform, but with a series of "stray hairs" sticking out above and below the buzzcut boundaries. In my experience the result is sonically not any better than the original. (No offense at all intended to Alex.)

 

Sometimes, though, you get lucky and you get a result more like what @fas42 shows above - a more natural-looking waveform that, in my experience, tends to actually sound better than the squashed original.

 

I like to use Izotope's declipper for this stuff, because it (A) scans the waveform and recommends a declipping threshold, and (B) it shows you a very helpful distribution graph of volumes/levels in the waveform, both before and after declipping.

 

Izotope's suggested threshold is often wrong, but when it's right, it makes the process super-easy. Even without that, the level-distibution graph can show you if (and where) the bulk of compression/limiting is concentrated in a certain volume range. That can help you set the threshold.

 

Finally, and most importantly, the post-declipping level-distribution graph can tell you much better than your own eyeballing of the waveform just how smoothly distributed the new levels are - in other words did you just do some declipping of a relatively small percentage of the peaks, or did you actually redistribute the levels of the track in a natural, and therefore sonically improving, way?

 

Oh - and Audacity and other apps do have actual dynamic expander plugins that theoretically can undo compression. However, I've tried them and found the results to be absolutely horrid: Either the expander doesn't mess up the sound but also doesn't do much expanding; or else it does do the desired level of expanding, but it completely messes up the sound, with the dreaded "breathing" or "pumping" effect that many expanders (both analogue and digital) are known for.

[sandyk]

7 minutes ago, tmtomh said:

Too often you get the result @sandyk posted a screenshot of earlier in the thread - same basic buzzcut brickwall waveform, but with a series of "stray hairs" sticking out above and below the buzzcut boundaries. In my experience the result is sonically not any better than the original. (No offense at all intended to Alex.)

 Why don't you try superimposing an  expanded timeline version of the before and after files using something like Sound Forge, or Audacity ?

 You will find that there are other improvements than just DeClipping as David L. reported in a previous post after listening to the uploaded before and after .wav files.

 

Quote

I compared the original and treated (SeeDeClip) files from Alex of Better be home soon.

The treated I thought was a winner, not night and day but, with impression of less congested sounding, the voice a little more separated and front, and the intro tending to bloom as in build a little more noticeably. That was matching volume by ear with the treated needing a boost of 2dB (2 notches on Volume)......

 

[gmgraves]

3 hours ago, fas42 said:

Hmmm, I slowed down a bit - had a heavy session with the dentist yesterday, and wasn't feeling brilliant for a while ...

 

Anyway, to get a feeling for what before and after waveforms can look like, here's sandyk's original version, attenuated by 7dB:

 

 

and the restored waveform, latest effort:

 

 

Pretty close to optimum, but I still want to play around with it a bit more; there's a slight volume pumping I want to eliminate.

 The volume pumping is always the big problem when trying to expand compressed commercial audio recordings. There is no way to know the compression rate and even more important, the attack and release parameters used by the record company to compress the audio. round trip compression can be pretty benign when those parameters are known. If you compress and expand using the same compression algorithm (or the same hardware compressor) you can be assured that the result will be free of pumping But when you don't know what the original characteristics were, it's always going to be a guessing game, and you're unlikely to hit it exactly. That's the reason why I think that a software compressor should have the reciprocal parameters built-in. 

 

BTW, I always thought that car stereos should come with a compression algorithm built-in and applicable at the listener's demand (IOW, one can turn it off and on at will).

[sandyk]

32 minutes ago, gmgraves said:

BTW, I always thought that car stereos should come with a compression algorithm built-in and applicable at the listener's demand (IOW, one can turn it off and on at will).

 

 Most car stereos need Decompression, not Compression, as the vast majority of FM stations use far too much compression in order to sound the loudest on the FM band, or still be heard when the car is doing 120kmh on  a Freeway.

Decompression/expansion  could be handy when listening to FM stereo when the car isn't in motion ?

[Audiophile Neuroscience]

1 hour ago, gmgraves said:

The volume pumping is always the big problem when trying to expand compressed commercial audio recordings.

 

I did a quick google but was unable to get a relevant hit for volume pumping. Anyone have time for a quick definition?

[mitchco]

volume pumping: https://www.gearslutz.com/board/409523-post7.html

 

As @gmgraves says, there is no way to know compression ratios. attack and release times, and the fact that (over) compression can be applied on individual tracks and at all stages of the recording, mixing and mastering process. More info, along with some sonic samples:

 

 

[Audiophile Neuroscience]

5 minutes ago, mitchco said:

volume pumping: https://www.gearslutz.com/board/409523-post7.html

 

As @gmgraves says, there is no way to know compression ratios. attack and release times, and the fact that (over) compression can be applied on individual tracks and at all stages of the recording, mixing and mastering process. More info, along with some sonic samples:

 

Thanks Mitcho, I will have a look at both links

[fas42]

Since I've got a couple of eagle-eyed recording engineers watching on, I'd better get this worked over track in pretty good shape before presenting it, - so, I might play with it a bit longer. Busy for a day or so, then I'll do another round of fiddling ...

[sandyk]

13 minutes ago, fas42 said:

Since I've got a couple of eagle-eyed recording engineers watching on, I'd better get this worked over track in pretty good shape before presenting it, - so, I might play with it a bit longer. Busy for a day or so, then I'll do another round of fiddling ...

 Hi Frank

 Although you are likely to be able to obtain worthwhile improvements with this particular track, it was far from a basket case,  with very little clipping in comparison with many modern recordings.

 Unfortunately, the results that you achieve may not lend themselves to using the exact same method for "correcting" the whole album , which many members would find quite acceptable as it is.

Unless we develop automated methods for Batch tasks, there will be far too much expenditure in time involved.

I believe that Graham Wilkinson from Cute Studios (I hope I got the name right ?) did a fine job in developing a program that was able to make many annoying recordings of the time more listenable.

Kind Regards

Alex

[fas42]

1 minute ago, sandyk said:

 Hi Frank

 Although you are likely to be able to obtain worthwhile improvements with this particular track, it was far from a basket case,  with very little clipping in comparison with many modern recordings.

 Unfortunately, the results that you achieve may not lend themselves to using the exact same method for "correcting" the whole album , which many members would find quite acceptable as it is.

Unless we develop automated methods for Batch tasks, there will be far too much expenditure in time involved.

I believe that Graham Wilkinson from Cute Studios (I hope I got the name right ?) did a fine job in developing a program that was able to make many annoying recordings of the time more listenable.

Kind Regards

Alex

 

No clipping, but quite severe compression. The improvements are major in terms of the quality of the instruments, the voice less so - what the compression does is cripple the sparkle and texture of the track, something which is important to me; so I'm happy to to expend some thought in reversing that.

 

In an album, if the sound engineer has some integrity he would tend to use the same processes on all the tracks, so that there is some consistency to the style of sound. So, a solution for one may set the pattern for all.

 

Of course automated methods is the answer, but the first step is manually, laboriously refining a procedure which really gets the job done - the learning curve is all-important, not the you beaut push the magic button and out pops a pristine track.

 

I looked at CeeDeClip about 6 years ago, and felt that better could be achieved. The latest version may be much improved, so I will take a closer look at it.

 

Cheers,

[sandyk]

1 minute ago, fas42 said:

In an album, if the sound engineer has some integrity he would tend to use the same processes on all the tracks, so that there is some consistency to the style of sound. So, a solution for one may set the pattern for all.

 Hi Frank

 You are unlikely to be able to do that with this particular series of 7 albums,(Best Audiophile Voices)  as they were sourced from many different albums, and remastered at a higher resolution before being converted to 16/44.1 again.

The originals would most likely have been engineered by different recording and mastering Engineers. 

 

Regards

Alex

[gmgraves]

19 hours ago, sandyk said:

 

 Most car stereos need Decompression, not Compression, as the vast majority of FM stations use far too much compression in order to sound the loudest on the FM band, or still be heard when the car is doing 120kmh on  a Freeway.

Decompression/expansion  could be handy when listening to FM stereo when the car isn't in motion ?

About broadcasts, I agree, but listening to classical CDs in the car, or one's own classical music files, soft passages get lost in the car/road noise. Compression is definitly needed in those instances.

[Audiophile Neuroscience]

2 hours ago, sandyk said:

I believe that Graham Wilkinson from Cute Studios (I hope I got the name right ?) did a fine job in developing a program that was able to make many annoying recordings of the time more listenable.

 

I would really welcome more input from him.

[fas42]

2 hours ago, sandyk said:

 Hi Frank

 You are unlikely to be able to do that with this particular series of 7 albums,(Best Audiophile Voices)  as they were sourced from many different albums, and remastered at a higher resolution before being converted to 16/44.1 again.

The originals would most likely have been engineered by different recording and mastering Engineers. 

 

Regards

Alex

 

And of course compilations are in a completely different basket - what really got me going in this exercise was a freebie CD, which had a bunch of theme songs, from TV shows. There was one in particular, which worked on the tiny speakers on the box - but played back on a fairly decent setup the sound was incredibly aggressive - a "take no prisoners" quality to it. Surely the musicians, a classic garage rock band, didn't sound like this in the studio? And indeed it had been given a quite simple, but heavy duty compression treatment - which could be reversed quite straightforwardly ... ahhh, real music !!!

 

The giveaway when reversing is weirdness in the sound, or volume pumping - all traces of that need to be excised, to call the exercise a success.

[fas42]

Just to say, I haven't given this away ... as usual for me, once I start getting into something I instinctively go deeper and deeper - so I have been sidetracked into looking into my old efforts, from years back, and checking what's happening out there in Net land...

 

The Better Be Home Soon track has been soft limited, a specific type of compression - I can go from the original to the mangled version from the Audiophile Voices album almost perfectly; the trick is now to go in the other direction, which is what I'm looking at at the moment.

[gmgraves]

On 10/9/2017 at 7:58 PM, sandyk said:

 

 Most car stereos need Decompression, not Compression, as the vast majority of FM stations use far too much compression in order to sound the loudest on the FM band, or still be heard when the car is doing 120kmh on  a Freeway.

Decompression/expansion  could be handy when listening to FM stereo when the car isn't in motion ?

 

Sorry, I already commented on that. My bad!

[witchdoctor]

On 10/6/2017 at 2:36 AM, Audiophile Neuroscience said:

Can Bad Recordings/ masterings be fixed and made to sound good?

 

Maybe a little creative EQ for each track?  Decompression/Declipping? Maybe a better system does it for you (realizing the potential within) ? Color me warm with sonic sunglasses?

 

This topic was inspired by posts in another thread some of which are quoted below to kick things off.

-----------------------------------------------------------------------------

 

 

 

 

 

 

I am using the Aftermaster Pro with bad recordings and yes, it does make them sound better:  This describes some of the trickle down tech used in the product  www.aftermasterpro.com

 

 

[Audiophile Neuroscience]

21 hours ago, witchdoctor said:

I am using the Aftermaster Pro with bad recordings and yes, it does make them sound better:  

 

I hear alarm bells. Demo tracks sound louder. Do you know what it actually does (that's a question not criticism) ?

 

https://www.head-fi.org/threads/the-aftermaster-pro-and-why-its-bullsh-t™.806030/

[witchdoctor]

19 minutes ago, Audiophile Neuroscience said:

 

I hear alarm bells. Demo tracks sound louder. Do you know what it actually does (that's a question not criticism) ?

 

https://www.head-fi.org/threads/the-aftermaster-pro-and-why-its-bullsh-t™.806030/

Its proprietary, they won't say how it works. I demoed it at www.aftermasterpro.com with the A/B tracks and bought one at amazon so I have a 30 days to return. So far I like what it does on hulu, old movies recorded in stereo (not 5.1) and lossy files through amazon music. I listened to some poorly recorded tracks by Jackie Wilson, Otis Redding, and the Four Seasons. I like it so far and have 2 more weeks to evaluate.  It's connected to my firetv via hdmi.

[CuteStudio]

On 10/11/2017 at 12:47 AM, Audiophile Neuroscience said:

 

I would really welcome more input from him.

 

Thank-you!

Caveat: The last time I suggested someone try my software on CA I was told I'd be banned if I did it again, so I've stayed off this thread and I'll be trying to keep any comments VERY non-product specific. Banning seems very popular these days in all forums about all subjects so it's impossible to have a robust conversation that would have been the norm 20 years ago, so I'll try to keep it very PC and wishy washy as if I didn't really care about the subject.

It's also a bit long as a post, but that also means I won't have to post for a while again too.

 

Expansion

Expanding waveforms is as noted very hit and miss. There are some major issues I found with with expanders:

a) The compressions can use multiple and variable equations and the 'look ahead' limiters can vary their effect on a peak-by-peak basis

b) When something is limited to a hard level you end up with the exact same digital number at the top of each peak, peaks that started off at different levels. No single expansion equation can change identical values to different levels unless you get back to guessing based on peak widths, but then the widths and slopes vary depending on where you look in terms of dB away from the peak.

c) it's very easy to add 'noise' - i.e. to make something sound far worse than it was in the first place. Very easy.

 

Measuring compression, limiting and clipping. 

Histograms can be very useful to show what expansion might work but even histogram shapes can be fooled. Usually a small 'bump' on the end of a histogram shows a super-compressed layer (limiter footprints) of about 1-5% of the signal containing a signal that wants to be 30-100% of FSD. However one track I saw had just flat tops, but these were slightly noisy flat tops and were dancing about in level, so if you believed the histogram and did an inflation of the squashed peaks you ended up with merely taller flat tops.

 

Histograms

Histograms will however reliably point out the amount of compression/limiting/clipping in a far more reliable way than any 'dynamic range' (peak/RMS) measure gives. I use the formula (IIRC) 1300 x samples_above_80% / Total_samples. This the gives a compression of 0 - 1 for very very dynamic tracks (even a constant sine wave is dynamic in this context) and 25 and up for some seriously damaged material (approaching a squarer wave which pushes samples to the waveform edges). In this way I get an instant view into how much compression was used, and usually how it was used. Often both a well populated 80-100% histogram is seen with a spike are seen, indicating both compression  and clipping. Clipping creates the spike in the histogram as there are an unusual number of samples sitting at the same spot on the extent. I also tend to use a log scale as the spikes are normally so tall they swamp the histogram so in practice a log vertical scale is much more useful to read.

 

Usually an 'honest clip' is easiest to 'fix', but mastering engineers are getting friendly with limiters more now so you end up with 'noisy' flat tops with rounded edges, some of the noise is just noise, sometimes it's a tiny signal that can be inflated. Any EQ after the clipping is also a problem, 'Black Ice' from ACDC went one better and had clipping that was then EQ'd (bass cut) and then clipped again. A few tracks are like this as well and you can see the flat slopes in the middle of the waveforms. Sometimes mastering people play the final track through an ancient analog compressor and resample it too which must I think add a D/A + compressor/limiter + A/D layer of distortion on top, this might explain some internal clipping one sometimes sees.

 

It's not just the big labels either BTW, the other day I was interested to see that Anna Nalick was self-producing her new album and her previous effort was quite well mastered considering, the new one though had a closed in LoFi sound so I looked up the histogram to reveal very weird and severe histogram (heavy and steadily rising all the way along when it should have been falling). It's not actually clipped, but the compression we pretty severe and the dynamic range (DR: Peak/RMS) was below 9dB which backed up the story. The same type of music from the 1980s has a DR of around 15-20dB. Such a shame, ruined on the way out of the door. I guess people forget that limiting is just as bad or worse than clipping in some instances.

 

So I find the histogram a very useful tool for analysis, not much gets to hide from it, the value distribution is fundamental to the chart.

 

The waveform graphs

Below is an Audacity snapshot of a typical not-too-badly mastered track of today. When you see and hear this it's in the 'modern' but 'not too tragically damaged' category.

 

This completely sums up the problem with mastering for me, because mastering engineers will line up to explain how it's actually a good master and that mastering is the complex, skilful business of stoic professionals. To be fair the mix and balance is very good, my only complaint is that I wanted the 'whiskers' on the loud bits too. I'm reliably told that no one else wants those and I get the impression that audiophiles like me whining about mastering quality are the problem, rather than the actual limiting/clipping which everyone wants. I've been told to use my ears and not to look at these graphs too, unaware that my ears led me to look at the graphs... and so the circle is complete.

 

The graph shows the problem from the HiFi view: this music has been squashed to fit into the medium: a mere 0.5 (3dB) headroom on each side would have given us a wonderful clean sound. Compressed yes, but clean. You can see the quieter parts are allowed to breathe, it would be so nice if the louder bits were allowed a breath too. Superficially it looks like all they needed to do is knock 3dB off the level and all would be fine, but there always seems to be a huge number of reasons why that's not possible or desirable, although they seemed to have managed it in the 1980s and 90s somehow.

 

 

'Come on, Come out' AFF and a million similar tracks..

 

I think this graph shows us why we have such a problem (note the 'we': it's our problem, no one else cares), each quiet section gives a tempting glimpse into how it could all sound, but doesn't. It's so close, just 3dB away from sounding really good, just that the loud bits are missing all of their peaks. About 70% of my library must be like this. So close, yet so far. Again and again. Year after year. Good money after bad...

The difference between a) clipping the medium, b) clipping a fraction below or c) carefully limiting when it's done thousands of times in the same short track is in my (and my ears) have almost the same effect: the loud bits sound distorted and constrained.

This has about 10,000 clips (the number varies on the threshold used, not all clips are flat and some will be the top of mildly cmpressed peaks), which affects about 90 sec of sound giving an average rate of 10000/90 = 111Hz.

 

This problem is now endemic and affects all digital formats and genres - not just pop and rock and I think it's something we'll have to live with for the next few decades. Apple's 'Mastered for iTunes' is far better - possibly because over half their introduction to highlights and forbids clipping, and possibly because in two places they reserve the right to reject it for sub-par quality.

http://images.apple.com/euro/itunes/mastered-for-itunes/docs/mastered_for_itunes.pdf

 

Some of the worst I have heard is Norah Jones 'The Fall' (unfixable, still can't get past the first minute), and the last 2 tracks on Elephunk from The Black Eyed Peas which has huge 250+ sample clips in it. Funnily enough the BEP tracks sound pretty good fixed, while declipping algorithms are constantly being advanced it's down to the actual damage regarding the ease of 'fixing' the waveform. 

 

Declipping

Flat tops have no information left in them, so there's a choice if either a) filling it in with a guess, b) filling it in from the other channel if that's unclipped or c) finding a bit of waveform nearly that might fit. 

 

Limited or 'rough' flat(ish) tops sometimes have a waveform compressed into them, so although the waveform is basically clipped, the clip is not completely flat so one can inflate the data which should be similar to the original (but still distorted of course).

 

In my view declipping works on a scale from 'badly' to 'very well' depending on the algorithm, settings and the track. It's always changing too, an update may declip better than an older version so keep your programs up to date.

 

Recommendations

This is what I and many others do, but you may find your own method is best.

 

Preview CDs. I love to listen to CDs mastered by people who keep off the limiters and use skill and judgement instead to attenuate and compress only as required. The odd few are still pretty good, the odd one exceptional (E.g check out Chlara: 'A Different Light'), diamonds in the sand. Any CDs that are damaged should in my view be taken back to the shop for a refund like you would any dud software.

BTW I'm not calling them 'silver discs' anymore as i recently bought a Charlotte Gainsbourg CD (IRM) that is completely black on both sides. Like pitch. Plays fine, but an interesting look!

 

Another way is to listen to everything in the car, but cars like the latest Audis have such good sound systems that it doesn't work so well anymore, so I'd advise an older car for this approach.

 

Buy older CDs and vinyl. CD is far better than vinyl in terms of the information it can hold and there's some superb stuff on CD, but the loudness war has meant it's rarely a HiFi format today in my experience (and listening) so that's why I recommend considering vinyl. Even Tesco has a vinyl section now so there most be a market of people fleeing digital limiting.

 

Also consider buying 'Mastered for iTunes' instead of CDs. Lookup the forums that discuss them, I just haven't see the same comments that CDs regularly get about our 'loudness' problem.

 

And the final rather controversial suggestion: Have a look at the type of waveform you are playing. Audacity works fine. I see the waveform of every single track I play (unless I have my eyes shut or am reading ZH instead) and have learned the correlation between the number of whiskers a waveform has left to the air, breadth and realism of the sound. My (minority) opinion is that HiFi today has a serious problem that starts with the way  CD/digital is mastered today. Plug in an original Dire Straits 'Sultans of Swing' CD, select 'Private Investigations' and listen to the sound. I have to do that now and again just to remind me just how good the CD format can sound.

[fas42]

4 hours ago, CuteStudio said:

 

Thank-you!

Caveat: The last time I suggested someone try my software on CA I was told I'd be banned if I did it again, so I've stayed off this thread and I'll be trying to keep any comments VERY non-product specific. Banning seems very popular these days in all forums about all subjects so it's impossible to have a robust conversation that would have been the norm 20 years ago, so I'll try to keep it very PC and wishy washy as if I didn't really care about the subject.

It's also a bit long as a post, but that also means I won't have to post for a while again too.

 

 

Thanks for posting your ideas on the matter! I haven't digested them as yet, but will do so shortly - and most likely respond.

[semente]

On 10/6/2017 at 7:36 AM, Audiophile Neuroscience said:

Can Bad Recordings/ masterings be fixed and made to sound good?

 

No.

[fas42]

7 minutes ago, semente said:

 

No.

 

The answer is vastly more than a simple No, or Yes. First of all, why is the recording "bad" - was the equipment used in terrible shape, of poor quality, and imposed a strong distortion signature on the end result; or was the recording gear of superb quality, the best one can buy, but the mastering "mangled" the raw inputs? There is a huge spectrum here, and the solutions will, or may vary per recording.