Hey guys,

I am working on restoring a lost-to-time recording, just for personal, hobby purposes.

This is a song that was recorded possibly off of a TV or the radio in the early 70s or late 60s. This "TV" effect was the first thing I noticed when I heard the recording, but I really just thought that it was unfixable, so didn't even try to tackle it.

I've done a lot of work already to straighten out horrible inconsistencies, drop outs, hum, etc, and after I fix some sibilant distortion, I'm looking at a fairly decent sounding recording (my standards for "decent" are probably way different and lower than most of people here!), but there's nothing I can think off to offset or minimise this very obvious "recorded by placing a microphone next to an analogue TV" sound.

I'm submitting the track here for you guys to have a look and maybe give some feedback - it's never gonna sound "amazing" or studio quality, but if I can minimise this effect, just to the point where it's not the first thing you notice upon hearing it, that would be like moving a mountain, because I've never known how to do it, or seen anyone else do it, plus I have more tracks from different tapes with a similar effect to different degrees, so it would be useful.

Here's a google drive link, I hope this does not violate any rules, it's a simple WAV, it was just too big to attach here.

[https://drive.google.com/drive/folde...H?usp=sharing]

To know how to reduce the effect, it's helpful to know how to create the effect in the first place, and I think with some EQ scooping, and a room reverb, you could create a similar effect. This recording doesn't really have any low-end at all, which certainly contributes to the problem (but doesn't cause it entirely, because boosting the low-end to absurdity doesn't mask the effect), and regarding reverb, I don't know if I can use any type of reverb reduction tools as this isn't dialogue in a corridor, it's a whole song. I think the use of such a tool would cause more harm than good.

I think another contributing factor is the disparity between the vocals and the backing, the level of the audio drops massively inbetween whenever the singer sings, somehow the vocal is just much louder, I am not sure if this was the original sound, or some effect created by the reproducing device. Perhaps a massive compression in dynamics would help, but again, that's quite a destructive procedure.

If anybody has any suggestions or advice, I'd love to hear it!

I listened to the supplied audio track. There is a lot more going on "mechanical wise" with the tape. It sounds exactly as you are describing. I will see what I can do and report back. I am working on a similar track for a client. I did get some work done using RX11 and the Wow and Flutter module. Lots of tapes from the 50s and 60s seem to have this problem. Best of luck in your quest...

Hey Tom,

Thanks for your reply - yeah, lots of problems, I know.

I spent about 3 hours yesterday working on the sibilant distortion, the "ch", "zh", "s", and "sh" sounds are somehow far too loud. I tried to strike a balance of not eliminating these sounds entirely, because being able to hear these sounds is a bit of a benefit for clarity, as often old muffled recordings don't have proper sibilance at all, to me having too much is better than having none, because it can be tamed.

I also tried to eliminate the "jarring" effect caused by whatever compression this went through. For example in between when the singer sings, the background noise slightly raises because the input signal momentarily gets lowered, and then when the vocal starts up again, at certain parts that split-second the compression takes to kick back in, results in the first part of a word being much louder than the rest of the sentence, at like 1:29 is a good example, that "Nam vypalo", the start of the "Nam" is very jarring.

I then analysed the spectrogram again and looked out for things like piano chords (where, there shouldn't be any kind of vibrato), and did see some "flutter"? I guess is the term, quick wobbles in pitch. The "wow & flutter" tool in iZotope is a bit of an auto-tune and can actually really mess with the intrinsic sound of the recording if over-used, so I double checked it had a good hang of what was flutter and what wasn't, because it gets it very wrong on some songs, but here it seemed to be smart enough, and so then I used it at sensitivity 1.5 for the entire recording, I don't believe it did anything significant to the vibrato in the vocals, which is the main thing to preserve when using pitch-leveling tools. I'm trying a sensitivity of 2.5 now as well to compare, but this is a very intense operation, so once I start it, I limit the CPU usage of iZotope through an external program so that it doesn't cause my computer to grind to a halt, and then I wait half an hour for the result.

After this, I can apply a small amount of compression just to limit any major differences in volume (because the volume does fluctuate a bit, but not in obvious ways I can see myself), and that gives me a result like this, once normalised to be at the max volume without clipping:

[https://drive.google.com/file/d/1Tyt...w?usp=sharing]

Obviously, you apply destructive things like compression at the very end, so the above is just for listening.

If I had no idea how to fix the "room microphone TV" effect, which I don't, I would probably be stopping here and calling it a day. I just wanted to show my reasoning and the result I can produce.

To me this is a fairly listenable result, it doesn't hurt your ears, there aren't major jumps or drops in volume, the sibilant distortion issue is gone, the sound is quite "full" for what it is, it's just still very obviously a TV recording. I am considering going back to before the compression, and filling out the low end some more. In testing, this added a low, rumbly kind of noise (one you feel instead of hear), but it made the recording sound warmer and less hollow, and made the "TV effect" maybe 20% less noticeable - so it might be a worthy trade-off. If I do this, I'll post it as well.

Just in case, here's a version of the file which is the same as the link in my very first post, except with sibilant distortion reduced, in case you want to carry out your tests on a slightly better version, as I believe the sibilant distortion reduction went very well:

[https://drive.google.com/file/d/1XU1...w?usp=sharing]

And here is the original version of the recording without any kind of EQ corrections or anything, (in theory, if you stumble upon a working method for sorting the "room effect", this might be the best way to test it), but it's also just a bad sounding recording, so if you have a method for turning the below file into "a bad recording without the TV effect", then the smartest thing for me to do would be to go back to square 1, apply that specific correction, then re-do everything else.

[https://drive.google.com/file/d/1LqK...w?usp=sharing]

Update: iZotope just finished the flutter reduction at 2.5 sensitivity, and as I expected, it dramatically changed the waveform of the singer's vibrato in a way I'd call unacceptable, it must be really hard for the algorithm to differentiate between flutter and a singer's vibrato, which is shame, but makes sense, as I bet most of their training data was 80s cassette tapes, not 60s reel-to-reel recordings.

I will listen to those later this afternoon and try some "things" to see if I can do anything. It is frustrating when you know there is more "hiding" if only you could get it to sound better. Keep up the GREAT work.

To my ear, the most offending artifact here is tape dropouts. There are hundreds of them in this recording, esp. during the intro/outro and the first chorus (0:52 - 0:53). Some manual work with Spectral Repair can bring them down like this:

Aleksey,

It was not very easy for me to see what you did, but after squinting long enough, I can see what you did. Yes, there are "micro-dropouts", to be honest, whilst they're noticeable, they weren't really on my radar, but seeing your work, it seems that maybe I can improve some of them, and the difference is quite remarkable, I just compared the original unedited file to yours. I found one that you did at around 0:52 seconds, and again, I didn't know that was possible!

However, I don't agree that it's the worst offender, to me that definitely seems to be the boomy "room reverb" effect, whilst you have brought up a valid problem with the audio (after all, the first step to fixing a recording, is to identify all it's problems), it was not really intended to be the subject of this forum thread. But I am still going to keep it in mind and add it to my to-do list.

I am currently trying to do some work with de-reverb tools, because I am sure that may be part of the puzzle, iZotope has one that I've never used before so I am trying to figure it out, and I am looking at installing maybe the Acon one, and I've heard a lot about Waves "Clarity Vx Pro" so am installing that as well (if it installs on my system...), to see what can yield the best result - as I believe none of them were made for mixed recordings, they're all meant for vocal tracks or single stems.

I agree. Tape drop outs can be "baked in" to a tape recording and there post processing is the only way. But it can also be due, in part or in full, to poor tape preparation, poor machine maintenance or adjustment, or all of the above. Expert work before and during digitisation can prevent many tape drop outs occurring in the first place.

I would be very careful, as these tools have only been created to remove reverb from speech, not music. I don't find that reverb in this record is objectionable.

Like so many of these audio "restoration" issues it really helps to know what is the actual problem and how it was caused in the first place.

The problem is the "mic in front of the speaker" thing adds in all the spectral non linearities of the mic, plus those of the speaker as well as other possible issues.

It helps to try and reduce or eliminate those non linearities with EQ, as you already knew. But it can be quite difficult to know exactly what those errors are, amidst all the action of the recorded performance. There used to be a software tool called HarBal which analyzed recordings for such spectral anomalies over a sufficiently long recording, displayed them and subject to operator fine tuning it corrected or at least reduced their severity. I've not used the tool but have heard results which can vary from great to not so great depending I guess on the source material.

I think there are such tools available nowadays but dont know what they are called. I'm sure someone here will know.

Aleksey,

I appreciate the points you made. I was well aware of the same thing (these tools are to fix badly recorded vocals, or maybe a single instrument, never are they advertised at being able to help with a full recording), but you don't know until you try.

I had no luck with the Waves one, and didn't get around to the Acon one, I figured if I gave the iZotope de-reverb a fair shot, that would give me a good indicator if it was worth pursuing other similar options. I played with the settings extensively, found where all the biting points are, etc, saw on the spectrogram what it did to the audio, and heard what kind of results it can produce. It's actually very smart software, and I have no doubt it can do a wonderful job on vocals maybe, but on a full recording, it's too much of a feat for the software to be able to do anything remotely smart.

timtape,

A quick google of "Harbal" and it immediately seems like an interesting piece of software that may have a good use-case - it doesn't immediately seem catered to restoration efforts, more like mixing/mastering, but that never stops me from taking a look if I can get my hands on it, It miht be very interesting. There are some bits of software out there which are so difficult to come by, or are incredibly expensive - for example an expander, the opposite of a compressor, I only have one such VST, and a harmonic balancing tool which does any kind of analysis is something I don't possess!

Back to Aleksey,

I believe you have iZotope and perhaps the ability to look at spectrograms, well here's the results of yesterdays work, I focused on eliminating the minor dropouts that you pointed out, and again, they're not all minor, some are quite noticeable.

I thought I would make this a bit more interesting, I am uploading the rxdoc for the work I did specifically with the dropouts, I don't actually know how it works, let me know if the project complains about missing a file, I'll send the exact file I used. if it works you should be able to trace through every single action I did, you can probably tell if I am doing a bad job or not in just a couple of minutes, only if it interests you. You can tell I got tired playing the game of luck with spectral repair and tried to save it for big drop-outs, nobody is perfect but there is some good work in there.

[https://drive.google.com/file/d/1GXA...sp=drive_link]

I used the spectral repair "pattern" mode for complicated areas, and the "partials + noise" for obvious gaps with clear harmonic content on both sides of the dropout. I use the "time-frequency selection tool" to target specific areas, because these tools are a bit like gambling, sometimes they will produce an amazing result, sometimes they will hallucinate or create harmonics that simply aren't there, or fudge nearby harmonics inexplicably, and a few times they produce a good looking result which isn't good sounding, like adding a random noise - but with trial and error, there was much success.

Where that failed, I used simple EQ to target areas that were minorly dampened, where the dropout hadn't fully split the recording, to iron them out - there were various times I couldn't fix a dropout to my liking, so I settled for just ironing it out so that it was less noticeable - I just can't accept the invented frequencies and "spectral shapes" that the spectral-repair can introduce into the recording.

Let me know what you think, I was flying through it by the end, I felt like I learned something, but obviously there's still lots to practice.

So overall, I am kinda done with the track, I am still applying finishing touches, but may post the final result in a few days - nothing is ever really final, but due to the "TV effect", it won't really be a satisfying result. It sounds fine in headphones but it'll always sound weird through speakers.

Then it's time to do some further research and try to send the recording to anyone that might be interested in it, fan sites, etc, usually this doesn't work out well.

For example I came across an instrumental orchestral recording done by Hungarian "Studio 11", I would never have identified it if the radio announcement wasn't also caught on the tape, naming the orchestra - I reached out to them via their website, and they said something along the lines of "We collaborated with the USSR to produce a string of instrumental works for them, but we don't have any of these recordings, or any information about these recordings". Great track, didn't seem to care much about having a piece of their own history made available to them...

I also came across a concert recording of the song Mediteranneo by Italian singer "Milva" from the mid 70s, did my best to try and find it, sure enough, doesn't seem to exist - reached out to the official website of the singer with the recording attached, asking if it's a novel find, if they want it cleaned up and straightened out a bit, not as a paid service or anything obviously, never heard back from them - likely an unmonitored inbox.

I could go on and on! Eventually I'm going to put together some sort of crude website just to post my findings and restorative work, I'll post lyrics, dates, and then people can come claim their songs...

Anyways thanks for all the feedback and advice guys, audio engineering and restoration is one of those things people rarely talk seriously about, because for some people it's their bread and butter and let's face it, if you have a knack for it, or some cool tricks up your sleeve, obviously you wouldn't want to give it away or shout it from the roof-tops. It's just a hobby to me so I don't see it that way I guess.

And for anyone curious, this is where I'm at a day later, some more dropouts fixed, and that's pretty much it. Maybe I've developed stockholm syndrome and like the sound of the track just because I've listened to it so many times now, but this sounds ok (through headphones).

[https://drive.google.com/file/d/1NxP...w?usp=sharing]

There was a noticeable problem (I don't know the word for it?) at 1min42 seconds where they say "DavayTE zhe budem", that "TE" was sounding very weird and pronounced, likely a badly timed mechanical issue with the tape. I sorted this by finding a similar portion elsewhere in the track, and using an equal-power fade to splice a better portion of audio, we're talking like less than 200ms here. That's a pretty uncommon thing that I don't think a lot of people know about or would dare to try, but it worked out quite well.

The funny thing is after all that work, the end result barely sounds any different compared to the original unedited audio, still just sounds like a microphone next to a TV, but I still feel that I removed certain barriers to listening by straightening out the issues I could find - it may be fair to say that I've maximised how good this recording can sound, within the context of it being a bad recording to begin with. So I wouldn't call it a victory, but this is why we practice on not-so-important audio, audio that is already forsaken, so that you don't lose sleep over it

It might be worth checking Aleksey's profile

jamesp,

LOL, had no idea, good thing I didn't say anything too bad about RX8 or anything adjacent

I've checked the RX-doc and your approach made perfect sense to me. Where the SNR was decent, you've amplified the existing signal with an EQ. Where the SNR was bad, you've applied Spectral Repair's Pattern (copy-paste) or resynthesis modes.

I could still hear that many dropouts were not addressed by you yet, like during the intro @3.0, 3.3, 3.4, 3.8, 3.9, 4.0, 5.2, 6.5 s.

Aleksey,

I wanted to say a thanks for your reply. Especially because when some random guy asks you to review an rx project file, go through the trouble of downloading an opening it, is actually not a trivial amount of work, and you're more than within your right to refuse to do that. So thanks for taking a look.

I am not in the industry, don't usually ever get a chance to speak to people who do the same thing, anything I know I basically either intuited or stumbled upon through trial and error - so to get feedback from someone that potentially knows what they're doing, is actually pretty big for me. Not only that, but to have a feedback that is somewhat positive, you have no idea what that does for my confidence in myself. When I tried using spectral repair on a small minorly dropped section, and had no luck, especially when the audio is still visibly mostly there, just quieter, I thought to grab the EQ and get messy, but it seemed like a hack, like something shameful and unscientific.

Knowing that what I did wasn't completely crazy, helps me to know that maybe I'm not wasting my time learning this stuff. To add to this, I only realised this after sending it, but that project file had maybe a minute or so of audio that I just completely missed, at/near the start. I have since gone back and also done the start of the recording. I think also it's possible there are times when a dropout is visible on the waveform, but due to my ears not being able to detect it, i deemed it skip-able. Audio doesn't win any awards for a pretty spectrogram, after all.

In the past I would have automated an EQ envelope using Reaper, but to do what I did in 2-4 hours in Izotope RX8, would have taken probably more than 10 times that in Reaper, due to lack of visual feedback. But in Izotope it's tricky for a different reason, because if you suddenly decide you don't like an edit you did 20 edits ago, but want to keep the other 19 you did since, you are in a real pickle. Also, the concept of markers completely falls apart as well as soon as you begin making non-full-spectrum "customised" selections, because the markers can't reflect that. EQ was the right tool for the job (as opposed to gain) since some dropouts are more full-spectrum whilst some are worse at higher frequencies, every time I used EQ, I eye-balled what I thought was an appropriate EQ curve for the job, and tested/tweaked each one a few times till it looked+sounded ok. One disappointment I had was that EQ will ignore non-full-spectrum selections, and gets applied as if you made a full-spectrum selection. Using the EQ tool would have been easier if not for this.

If you really are associated with Izotope, I want to say you guys have done great work, I think the only stuff better than yours must be that Cedar stuff, but their products aren't available to the public, so who cares. By the way, I know you didn't ask for me to list my complaints with RX8, but if you happen to know about any of the above things (for example, are they fixed/improved in a later version of RX, I would find that information very valuable).

Thanks again to everybody for the replies.

Starting from RX 9, there is a Restore feature in the undo history to return to an arbitrary previous undo step for the current selection. Prior to RX 9, this could be done manually, via Copy + Paste-Replace.

The EQ, like most other RX modules, should respect the bandwidth of your selection. For very short selections, there may be some considerable transition band in the crossovers.

Anyone can buy the Cedar hardware/software but it isn't cheap and it is normally only available through proper professional audio suppliers.