The SeeDeClip4 Declipping Music Server

[CuteStudio]

@CuteStudio
There’s a discussion about your software on Roon forum. Brian from Roon made a recommendation I’d like to know your thought on his upsampling idea.

https://community.roonlabs.com/t/dsp-engine-de-clipping-and-or-dynamic-expander/33424/13

I think the attenuate + upsample idea (i.e. create some headroom and digitally filter it to knock the corners off) is not such a bad idea - I suspect this is what most CD players actually do internally. For tracks with lots of fine clips this would probably be a good plan, but the majority seem to have some decent 100+ sample clips - even tracks that don't look too bad.

This track I got from Sarah McLachlan's 'Afterglow' CD is pretty good all in all (It actually just scores 'Ace' with mild compression, about 1300 clips/limits and a reasonable DR of over 14dB). It does however have a 44 sample clip as illustrated in the studio page here:

In the big graph at the top in the middle you can see - the green line is where the original bumps along.
Note also the subtle 'notch' in the 'after' histogram, which shows my expansion algorithm needs further finessing. Incidentally this is another argument against batch processing - converting thousands of tracks takes time which is why it's a streamer - it just does the ones I listen to with the latest algorithm.

The difference between hiding clips under the rug (attenuate/filter) and fixing them is perhaps like the difference between a good MP3 and a good lossless track.
If you can find an unclipped original CD - then compare it to the 'remastered' album CD - you get perfect pre and post mangling listening test!!

The result I hear is a little more 'snap' and 'bite' and a little more air and space. Also remember some tracks declip better than others - The Black-Eyed Peas 'Elephunk' has some seriously damaged tracks (hip-hop is popular for this) but declips pretty well and is a good listen - but not all are, the art of the declipper is to handle different tracks as every one is different - some are a revelation, others you wouldn't notice.

 

[CuteStudio]

If you want to make money on your program, you may want to reconsider offering it as a VST plugin for a price. Virtually all DAW's and software media players, including consumer players, support the VST plugin model. KVR Audio is the largest marketplace of plugins, literally thousands of them: https://www.kvraudio.com/ And offers a platform for developers to sell their plugins. If your tech could support streaming services or simply real time declipping and offered as a plugin for I don't know $20 or whatever, your dream of huge amounts of cash could come true. Good luck!

Thanks for the input, when time and money permit I'll see if I can get a VST plugin development rig sorted out and see what it would involve.
Cash is just a means to pay for the HiFi, they say you can't take money with you but you might be able to take some drive units or a nice triode with you

 

[dallasjustice]

https://www.audiosciencereview.com/forum/index.php?threads/perfect-declipper.2209/

 

[CuteStudio]

https://www.audiosciencereview.com/forum/index.php?threads/perfect-declipper.2209/

Thanks for the link, nice demo, I'm impressed he managed to find some 1980s music that was clipped - not sure I've managed that yet

I like the sliders with their percentages, it illustrates the slightly ad-hoc nature of the task which I've been trying to apply some heuristics to. I also liked the real-time displays too, although the video fails to convey the sheer magnitude of many of the clips as yawning flat tops hanging around 50-250 samples - something with a decent bass thump is usually the worse.

Your comment about €100 being a lot for software is interesting and a common view, many people won't blink when spending €400 on a USB cable but think that somehow software has no instrinsic value. Just 2 hours of a software engineers time easily swallows that €100. There's about 1691 Git commits in SeeDeClip4, previous versions were not in Git and there's probably the same again before, each commit representing an average of 8 hours work, so about 28,656 hours work at €50 per hour is around $1,432,800 worth of R&D work in SeeDeClip4 available for £20 (Or free if you don't listen to the last half of the track LOL).

I'd guess the perfectdeclipper R&D is around $0.5M to $1M (minimum, as it fits into VST rather than being a music streaming system itself) so for 1/4 the price of a 'audiophile' cable you've got quite a bargain there .
These simple economics are the reason that cable sales are big business and software to actually improve the source is fairly rare. Decades of fake news HiFi marketing has created the widespread perception that someone stuffing cable into a hosepipe in his garage is creating more value than tens of thousands of man hours in software.

My mindset is more Steve Jobs than Mr SmallCogInSomeoneElsesMachine but without his marketing brilliance. His Lisa and Mac were before their time as is the web-app concept used in SeeDeClip4, and if I had the resources I'd be moving on to:

• Self contained hardware music server and slave boxes powered by SeeDeClip4
• A line of Class A 'superTriode' SET tube amps
• A deal with KEF or Tannoy (or similar class speaker manufacturer)
• A deal to stream HiFi tracks from selected artists with an emphasis on quality and content (i.e. this: http://www.cutestudio.net/data/SeeDeClip4/music would be an iTunes/Spotify type page rather than a 'try these' page)
• Sensibly priced cables to repel the cable sharks.
• To present an end-to-end system with some outlet 'RealFi' concessions and shops for people to buy and browse some decent audio gear.

The market is there, the technology is here, the manufacture costs are not too large, but HiFi is a very self destructive industry so it would require a firm like Apple (or even Microsoft who need a shot in the arm!) itself to decide to carve out a 'new' segment like Steve did back in the day when his ideas and technology finally meshed together with the iMac G3 in 1998.

It would be quite possible to bring quality back and make good HiFi trendy and sought after again IMO, but I'm still waiting for the call of an interested billionaire investor to discuss strategy. Tesla is burning $16m per day, for less than a week of that I'd have an 'Apple for HiFi' up, running and kicking

 

[Thomas savage]

I’m encouraged by these ‘declipping’ products , a great deal of what you find in home Audio seem to be solutions looking for problems but not so in these cases ..

 

[March Audio]

This is a real problem but the cynic in me asks is it a real solution? As very eloquently said elsewhere, "are we putting lipstick on a pig".

However, rather than dismissing it out of hand I would want to perform some controlled tests. Take an unclipped file, increase level in a daw to clip the file, run the declipping and see how close it gets to the original.

 

[dallasjustice]

I know where you are coming from and I'm not a fan of audio tweaks. However, I generally don't accept the market value theory of a product which is based solely on the time or effort put into making the product. It's flawed economic theory which has been disproven.

The truth is that value is subjective. Speaking strictly as a consumer, I think the value of this product could be perceived to be very high, IF it could be easily demonstrated and deployed in a more universal way. I think linux is a very limited platform. Converting static files is also unappealing to folks who stream music (most everyone now). Btw, the 100 euros I spent on the declipper software is a bargain for me.

I think the .vst plugin idea is a good one. I also think that partnering up with Roon or Jriver could be a good way to go. Sure you wouldn't get paid much per user, but you'd make it up on the volume.

Thanks for the link, nice demo, I'm impressed he managed to find some 1980s music that was clipped - not sure I've managed that yet

I like the sliders with their percentages, it illustrates the slightly ad-hoc nature of the task which I've been trying to apply some heuristics to. I also liked the real-time displays too, although the video fails to convey the sheer magnitude of many of the clips as yawning flat tops hanging around 50-250 samples - something with a decent bass thump is usually the worse.

Your comment about €100 being a lot for software is interesting and a common view, many people won't blink when spending €400 on a USB cable but think that somehow software has no instrinsic value. Just 2 hours of a software engineers time easily swallows that €100. There's about 1691 Git commits in SeeDeClip4, previous versions were not in Git and there's probably the same again before, each commit representing an average of 8 hours work, so about 28,656 hours work at €50 per hour is around $1,432,800 worth of R&D work in SeeDeClip4 available for £20 (Or free if you don't listen to the last half of the track LOL).

I'd guess the perfectdeclipper R&D is around $0.5M to $1M (minimum, as it fits into VST rather than being a music streaming system itself) so for 1/4 the price of a 'audiophile' cable you've got quite a bargain there .
These simple economics are the reason that cable sales are big business and software to actually improve the source is fairly rare. Decades of fake news HiFi marketing has created the widespread perception that someone stuffing cable into a hosepipe in his garage is creating more value than tens of thousands of man hours in software.

My mindset is more Steve Jobs than Mr SmallCogInSomeoneElsesMachine but without his marketing brilliance. His Lisa and Mac were before their time as is the web-app concept used in SeeDeClip4, and if I had the resources I'd be moving on to:

• Self contained hardware music server and slave boxes powered by SeeDeClip4
• A line of Class A 'superTriode' SET tube amps
• A deal with KEF or Tannoy (or similar class speaker manufacturer)
• A deal to stream HiFi tracks from selected artists with an emphasis on quality and content (i.e. this: http://www.cutestudio.net/data/SeeDeClip4/music would be an iTunes/Spotify type page rather than a 'try these' page)
• Sensibly priced cables to repel the cable sharks.
• To present an end-to-end system with some outlet 'RealFi' concessions and shops for people to buy and browse some decent audio gear.

The market is there, the technology is here, the manufacture costs are not too large, but HiFi is a very self destructive industry so it would require a firm like Apple (or even Microsoft who need a shot in the arm!) itself to decide to carve out a 'new' segment like Steve did back in the day when his ideas and technology finally meshed together with the iMac G3 in 1998.

It would be quite possible to bring quality back and make good HiFi trendy and sought after again IMO, but I'm still waiting for the call of an interested billionaire investor to discuss strategy. Tesla is burning $16m per day, for less than a week of that I'd have an 'Apple for HiFi' up, running and kicking

 

[dallasjustice]

"are we putting lipstick on a pig".

I'd say it's more like putting Bar-B-Que sauce on mediocre BBQ. It makes it go down a lot better. OTOH, if you have great smoked meats, you'd never put sauce on it. At least, we don't do that here in Texas.

 

[fas42]

This is a real problem but the cynic in me asks is it a real solution? As very eloquently said elsewhere, "are we putting lipstick on a pig".

However, rather than dismissing it out of hand I would want to perform some controlled tests. Take an unclipped file, increase level in a daw to clip the file, run the declipping and see how close it gets to the original.

Software reconstruction, "fixing" of poorly mastered material is certainly a real solution - in fact, down the track I see it as a cornerstone of the whole audio industry - as CuteStudio points out, people want to pay nothing for software, and staggering amounts for a bit of wire - talk about arse over tit ... . But the truth is, people do want to believe in magic solutions, the instant, perfect, fixit, bandaid, that makes it all better, without thinking about things for more than a split second ...

The movie industry has got this under control - they can restore old movies in terrible shape, and make them gleam like a brand new car ... just throw enough processing power at the job, and out pops a good "product".

 

[March Audio]

The subjective result (however right or wrong) could be more palattable than the straight clipped version.

 

[Wombat]

I would like to see a stand-alone product that can be used with existing analogue and/or digital systems.

 

[CuteStudio]

This is a real problem but the cynic in me asks is it a real solution? As very eloquently said elsewhere, "are we putting lipstick on a pig".

However, rather than dismissing it out of hand I would want to perform some controlled tests. Take an unclipped file, increase level in a daw to clip the file, run the declipping and see how close it gets to the original.

The best controlled test is to try it on real clipped music. Use a DAW and you'd not get close to the range and inventiveness of wrecking a good master. Some clip it, re-eq and then clip it again (AC/DC Black Ice is a good example of this), others clip it straight, others clip it with noise, others run it through a DAC into an old analog compressor and back into an ADC (this is disturbingly popular and 'trendy') and others will use a simple limiter. For every album there's a different way to clip. Sometimes they even clip the -ve going side differently that the +ve going sides - it's more popular than you'd think.

So the answer to controlled testing (as I mentioned above) is to by an original album and a remastered version. Then compare the two: it's a perfect test because we're really trying to get back to the quality of a 1980s/90s CD master from a modern mastering disaster. You can use the free bits of the SeeDeClip4 to do this.

Keep checking back though as every time I get a chance to work on the declipping again (it goes in cycles in the mass of other jobs) it gets better. For instance the declipping it does now is significantly better than it was doing a year ago. The improvement of software is why the update it easy to use and why it streams (although it can batch convert or 'export').

 

[CuteStudio]

I would like to see a stand-alone product that can be used with existing analogue and/or digital systems.

Ideally it would be a box you plug XLR (or our rubbish HiFi RCAs LOL) in and out of. Problems with this are:

1. You'll need another ADC and DAC in there with all the problems this entails for quality
2. Latency. It can take 5 seconds on a decent computer to scan and fix a track. Probably not acceptable for any analog path.
3. It's nice if you have the whole track to work with, an average of 4 minute delays may also be a problem.

The best (and therefore only - the job is difficult enough anyway) method is to edit the original digital data before it gets streamed, it simply means changing your whole system over to SeeDeClip4 to give you the full benefits - but you knew I'd say that didn't you .

You also have the option to export playlists to WAV, Flac (and MP3) so you can actually burn a CD or memory stick with and use any existing kit.

 

[Thomas savage]

What about giving it to the streaming service before they send the music out to the consumer?

 

[CuteStudio]

Speaking strictly as a consumer, I think the value of this product could be perceived to be very high, IF it could be easily demonstrated and deployed in a more universal way. I think linux is a very limited platform.
Converting static files is also unappealing to folks who stream music (most everyone now).

If you have music files on your PC it should only be a matter of minutes to be playing then from SeeDeClip4, did you try this - and which step made you give up? It would be very useful feedback as I'm in the worst position to know as obviously I get it working. This is why people should never test their own software

SeeDeClip4 on Windows 10 is the most popular platform - if there's something on the website that implies it only works on Linux please let me know!
I run it on OS-X too but the installer needs updating for that, so currently Windows (XP to Win10) and Linux and ARM Linux (Raspberry Pi). The Pi is getting popular for audio slave devices though - it's so cheap, quiet, low power and powerful it would be rude not to.. also a number of i2s DACs plug right into it now.

It can both stream or export to static files, I agree that static files are a pain because when the declipper gets an improvement I don't want to wait ages to re-convert them all again, so yes, definitely streaming is the way forward!

 

[CuteStudio]

What about giving it to the streaming service before they send it out to the consumer?

Most streamers only stream MP3s so would they be interested in quality? iTunes has 'Mastered for iTunes' where they forbid clipping, but that's mainly clipping of the medium - a lower magnitude signal with clipping may still get through, they've done it by tackling the problem head on by saying: "Master it properly for us to label it this way".

Is Tidal the only lossless streamer? Expensive and how compressed and clipped is JayZ's output (I have none- genuine question!).

For quality, economy and speed reasons I find local home-wide streaming to be the best of all worlds but it's quite likely that I'm in a minority! I export to a memory stick for the car.
The extra kit over internet streaming is only a PC or Mac which I have anyway..

 

[March Audio]

The best controlled test is to try it on real clipped music. Use a DAW and you'd not get close to the range and inventiveness of wrecking a good master. Some clip it, re-eq and then clip it again (AC/DC Black Ice is a good example of this), others clip it straight, others clip it with noise, others run it through a DAC into an old analog compressor and back into an ADC (this is disturbingly popular and 'trendy') and others will use a simple limiter. For every album there's a different way to clip. Sometimes they even clip the -ve going side differently that the +ve going sides - it's more popular than you'd think.

So the answer to controlled testing (as I mentioned above) is to by an original album and a remastered version. Then compare the two: it's a perfect test because we're really trying to get back to the quality of a 1980s/90s CD master from a modern mastering disaster. You can use the free bits of the SeeDeClip4 to do this.

Keep checking back though as every time I get a chance to work on the declipping again (it goes in cycles in the mass of other jobs) it gets better. For instance the declipping it does now is significantly better than it was doing a year ago. The improvement of software is why the update it easy to use and why it streams (although it can batch convert or 'export').

Well that depends if you ars trying to re-create the original unclipped waveform or just try to make it sound more palattable. There is no way you can unravel the list of processing you cite above in any accurate way. Compariing an 80s master to a modern remaster is faux for the same reasons; you simply dont know what they did. Was it just compressed and clipped or did they re-eq?

So Im afraid Im a little unconvinced. I can see you might be able to make something sound more pleasant if the output hasn't essentially been turned into square waves, but I dont think you can undo the liteny of damage created by bad recording or mastering.

 

[March Audio]

Most streamers only stream MP3s so would they be interested in quality? iTunes has 'Mastered for iTunes' where they forbid clipping, but that's mainly clipping of the medium - a lower magnitude signal with clipping may still get through, they've done it by tackling the problem head on by saying: "Master it properly for us to label it this way".

Is Tidal the only lossless streamer? Expensive and how compressed and clipped is JayZ's output (I have none- genuine question!).

For quality, economy and speed reasons I find local home-wide streaming to be the best of all worlds but it's quite likely that I'm in a minority! I export to a memory stick for the car.
The extra kit over internet streaming is only a PC or Mac which I have anyway..

Tidal has a "hifi" option. Where if the file supplied from the record company is the same as the cd, it sounds the same as the cd.

 

[CuteStudio]

Well that depends if you ars trying to re-create the original unclipped waveform or just try to make it sound more palattable. There is no way you can unravel the list of processing you cite above in any accurate way. Compariing an 80s master to a modern remaster is faux for the same reasons; you simply dont know what they did. Was it just compressed and clipped or did they re-eq?

Hmm, you asked: "I would want to perform some controlled tests"

In my opinion the ultimate controlled test is to compare an original and a repaired remastered.
Everything in your post I quote here is surely covered by that test - is it not?

So Im afraid Im a little unconvinced. I can see you might be able to make something sound more pleasant if the output hasn't essentially been turned into square waves, but I dont think you can undo the liteny of damage created by bad recording or mastering.

The purpose of this thread is not to convince! I'm just saying here that it exists and the aim of the software. I'm glad you tried it out though - thanks!

I appreciate your useful feedback, could you just note here:

• What system (OS) did you download it onto?
• How did you play the tracks - via the browser or a Raspberry Pi slave? 16 or 24bit?
• What tracks did you listen to? - just notable ones!
• Which tracks were not convincing?

Thanks in advance!!

 

[March Audio]

Hi Cute,

Im sorry but I havent tried it out yet, I have been away from home. Please dont misinterpret anything I have said here, I am just trying to get a handle on the objectives of the software. I can see it could make some clipped material sound more palatable by maybe reducing objectionable harmonics - thats useful, but comparing to a remaster is only a valid as a personal subjective opinion. Thats great, if any individual likes the result thats fine, howver its not a controlled experiment to see how the signals are reconstructed. The implication is that the software can repair. My view is that its not possible to undo the many and varied problems or changes that could have been made by a modern remastering without knowledge of what has been done. It could certainly change the sound however, and you may like the result.

I would certainly like to give it a try and will when I get time. You never know I may like the result