Tools for removing background audio buzzing sound from VHS transfer?

[Wanderer]

Hello all,

I have some VHS transfers where the audio is too low whenever people were too far away from the camera when it was recorded. When I boost the audio, there is a loud buzzing or hissing sound throughout. I've learned how to reduce it with Noise Reduction in Audacity, but it does seem to result in a little damage to the rest of the audio(it's possible I just need to adjust the settings better). I'm wondering if there is a better tool that the more experienced prefer. It helps if it's user friendly because I'm very much a novice at all of this, but I'm open to learning if the best tool isn't. I've attached an example of the background noise. Thanks in advance to all for any advice.

[netmask56]

Possibly an Earth loop - see https://www.jensen-transformers.com/product/ci-2rr/

[Wanderer]

Possibly an Earth loop - see https://www.jensen-transformers.com/product/ci-2rr/

Thank you for the suggestion. I'm not able to redo the capture, as I had someone else do that, so I'm looking for a software solution that would help me fix the audio in the recording I already have. Some kind of app with smart noise reduction.

[davexnet]

Izotope RX 11, free trial available.
Spectral Denoise and/or Dialog Isolate

[pandy]

Spectral subtraction and your ear are enough to remove most of the problems from your recording but proper methodology is key to success.

Key points are:

- use relatively small FFT (somewhere around 1024..4096 points), higher FFT can lead to more spectral leaks and as such various chirps
- choose proper sample for noise profile acquisition - it should be free from complex tonal signals - search for relatively short time frame where only noise is - present, no other sound especially things like vocals or music is present
- perform processing in chunks as in time noise profile may change in time so one acquired at the beginning of the recording may be different than noise profile in middle or at the end
- it is useful to search for optimal threshold level to select inverted processing i.e. removing signal while keeping noise - search for such threshold where noise is unmodulated by useful signal
- try to perform relatively mild noise removal strength - not go over 20dB - you can always repeat process multiple time to increase noise removal efficiency.

Btw i have impression that your source is largely mono - so i would at the beginning of processing combine both channels to one and as such gain approx 3dB better signal to noise ratio, also i would avoid increase recording level before noise removal, as your sound has practically no useful spectrum above 12kHz i would also apply bandpass filter - from 150Hz to 12000Hz , removing everything bellow 150Hz will cut eventual 60Hz power line hum and it's 2nd harmonic at the 120Hz.

[Barrythecrab]

Cyberlink AudioDirector can single out the noise and mitigate it.

[davexnet]

Updated Rx 11 to a revision put out by Izotope, just two days after the initial release.
Here's another go using the updated Dialog Isolate. Seems to be increased sound quality compared to RX10.

[Wanderer]

Thanks so much to all of you for the suggestions and advice. I have a lot to explore now.

Pandy to be honest much of what you've said here is over my head, I'm going to have to go learn some terms and measures, but thanks for expounding all that.


Davexnet honestly both of those samples sound pretty good to me, thanks for doing that. I'm going to go watch some tutorials now and see if I can get it there myself. Hopefully the Izotope trial will last until I'm done with all the VHS transfers

[davexnet]

Hopefully the Izotope trial will last until I'm done with all the VHS transfers

I believe it's full functionality for 10 days. Good luck with it !

[Wanderer]

Hopefully the Izotope trial will last until I'm done with all the VHS transfers

I believe it's full functionality for 10 days. Good luck with it !

In that case, I guess I better wait until my vacation to make sure I don't run out of days!

[forart.it]

RX11 and similar apps are obviously the easyest (but expensive) way to go.

For those who like experimenting, I've collected a quite-large list of open source software that *may* let you achieve the same (or even better) results:

HyMPS project \ AUDIO \ Treatments

Enjoy !

[bar72]

Cool Edit Pro's Noise Reduction. I've attached the reduction profile if you want to try it.

buzzing example-cooledit.wav

cool-nr-profile.zip

[greymalkin]

Cool Edit Pro!! You just took me back in time 20 years .

What do you think of the built in audio tools with the free DaVinci Resolve? I've used the Voice Isolation option there and it was a quick way to get rid of background noise as long as you don't turn it to 11 :P.

[greymalkin]

Something else that I find works surprisingly well for a 1 click solution (OK maybe 3-4 clicks) is the new-ish "AI" Noise Reduction that you can install for Audacity.

If you have a newer version of Audacity go to Effect - Get AI Effects and download/install the package. After install it will give you instructions for how to ensure the module for the ai effects is turned on.

Then restart Audacity, select the whole audio track and go to Effect - OpenVino AI Effects - OpenVino Noise Suppression. if you click Advanced in the AI Noise Suppression box you can enter different values to do light to heavy noise reduction.

I just want to re-iterate the suggestions above will likely give a better result! but if you want a simple, "good enough" solution (that's for you to decide! I have no idea how it will turn out)...give this a try.

[timtape]

I guess it's stating the obvious but recording the live audio properly in the first place is the tried and true method. Just using a proper directional mic, close to the subjects at time of shooting can make a big difference. There's only so much improvement that can be done in post.

As for denoisers including the latest so called AI types, it's rare that their results are ever compared not just with the poor recording they are trying to improve but compared to the decent recording that could have been made in the first place, if the recordist had known what they were doing.

[clashradio]

Spectral subtraction and your ear are enough to remove most of the problems from your recording but proper methodology is key to success.

-use relatively small FFT (somewhere around 1024..4096 points), higher FFT can lead to more spectral leaks and as such various chirps

I'm using RX9 Spectral De Noise with a vhs project. I've got the buzzing and most of the general background noise removed, but still get the occasional chirp. I can't find the "FFT" you mention. What plug-in are you using to help control chirps?

[timtape]

Spectral subtraction and your ear are enough to remove most of the problems from your recording but proper methodology is key to success.

It might seem otherwise but no post processing can truly remove that constant hissing, to unmask the pristine speech underneath. In Wanderer's sample (sounds like the voice of Larry Odham the quad video man and country guitarist)) every word is already fully intelligible with no Denoising. So any Denoising will not be to increase intelligibility but for 'cosmetic' reasons.

Below is a sample I made up where the noise predominates and masks most of the speech. I've not found any Denoise software than come even close to making this speech intelligible.

...Btw i have impression that your source is largely mono - so i would at the beginning of processing combine both channels to one and as such gain approx 3dB better signal to noise ratio...

Yes this is a good technique but only works where the wanted audio is common to both stereo channels and the noise to be reduced is not common. This doesnt apply here, and even if it did the reduction would only apply to the much lower noise (common to both channels).

Wanderer's is a HiFi stereo track (note the occasional click in one but not the other channel) but again the hissing is common to both channels. We'd be looking for a stereo track where the noise we wanted to reduce was not common to both channels, such as a mono linear recording played back on a stereo linear capable machine (stereo playback head). And that would only be of benefit if the playback head was perfectly aligned to the recorded linear track's azimuth, when summed to mono.

,,,also i would avoid increase recording level before noise removal, as your sound has practically no useful spectrum above 12kHz i would also apply bandpass filter - from 150Hz to 12000Hz , removing everything bellow 150Hz will cut eventual 60Hz power line hum and it's 2nd harmonic at the 120Hz.

Yes that's been a proven technique for many decades but this is not the best example for it. Here the speech bandwidth extends up to about the same as the captured hiss. We could lower the LPF to say 6 kHz but it would reduce both hiss and speech in that part of the spectrum.

[timtape]

Spectral subtraction and your ear are enough to remove most of the problems from your recording but proper methodology is key to success.

It might seem otherwise but no post processing can truly remove that constant hissing, to unmask the pristine speech underneath. In Wanderer's sample (sounds like the voice of Larry Odham the quad video man and country guitarist)) every word is already fully intelligible with no Denoising. So any Denoising will not be to increase intelligibility but for 'cosmetic' reasons. The further problem is that so often these days where Denoisers are so cheap and easily available, carelessly aggressive Denoising actually reduces speech intelligibility. Just because the noise seems lower doesnt mean the speech is made clearer. The speech may now be harder to understand.

...Btw i have impression that your source is largely mono - so i would at the beginning of processing combine both channels to one and as such gain approx 3dB better signal to noise ratio...

Yes this is a good technique but only works where the wanted audio is common to both stereo channels and the noise to be reduced is not common. This doesnt apply here, and even if it did the reduction would only apply to the much lower noise (not common to both channels).

Wanderer's is a HiFi stereo track (note the occasional click in one but not the other channel) but again the hissing is common to both channels. We'd be looking for a stereo track where the noise we wanted to reduce was not common to both channels, such as in the case of a mono linear recording played back on a stereo linear capable machine (stereo playback head). And that would only be of benefit if the playback head was perfectly aligned to the recorded linear track's azimuth, when summed to mono.

,,,also i would avoid increase recording level before noise removal, as your sound has practically no useful spectrum above 12kHz i would also apply bandpass filter - from 150Hz to 12000Hz , removing everything bellow 150Hz will cut eventual 60Hz power line hum and it's 2nd harmonic at the 120Hz.

Yes that's been a proven technique for many decades but this is not the best example for it. Here the speech bandwidth extends up to about the same as the captured hiss. We could use a LPF below say 6 kHz but it would reduce both hiss and speech in that part of the spectrum.

Wanderer, re the sample you uploaded, is the sample from the original camera tape? If not and it's a dub, possibly at least some of the noise was accidentally added when dubbing. A clean recapture from the camera tape would bypass that added noise.

[pandy]

I'm using RX9 Spectral De Noise with a vhs project. I've got the buzzing and most of the general background noise removed, but still get the occasional chirp. I can't find the "FFT" you mention. What plug-in are you using to help control chirps?

I use plain Audition 3.0 build 7283.0 - it has very useful feature -"keep only noise" - it literally inverting functionality and remove signal leaving just noise - you can search for optimal noise removing level - simply set reduction level to point where noise is NO longer modulated by signal - this is way how you avoid chirp (chirp i.e. spectral leak where noise is removed and also part of spectrum of useful signal).

[pandy]

It might seem otherwise but no post processing can truly remove that constant hissing, to unmask the pristine speech underneath. In Wanderer's sample (sounds like the voice of Larry Odham the quad video man and country guitarist)) every word is already fully intelligible with no Denoising. So any Denoising will not be to increase intelligibility but for 'cosmetic' reasons. The further problem is that so often these days where Denoisers are so cheap and easily available, carelessly aggressive Denoising actually reduces speech intelligibility. Just because the noise seems lower doesnt mean the speech is made clearer. The speech may now be harder to understand.

Technically if your noise is uncorrelated with signal then full separation (noise removal) is possible.

Yes this is a good technique but only works where the wanted audio is common to both stereo channels and the noise to be reduced is not common. This doesnt apply here, and even if it did the reduction would only apply to the much lower noise (not common to both channels).

Wanderer's is a HiFi stereo track (note the occasional click in one but not the other channel) but again the hissing is common to both channels. We'd be looking for a stereo track where the noise we wanted to reduce was not common to both channels, such as in the case of a mono linear recording played back on a stereo linear capable machine (stereo playback head). And that would only be of benefit if the playback head was perfectly aligned to the recorded linear track's azimuth, when summed to mono.

It help as i've wrote already for mono signal sampled as stereo and noise is uncorrelated between both channels.

Yes that's been a proven technique for many decades but this is not the best example for it. Here the speech bandwidth extends up to about the same as the captured hiss. We could use a LPF below say 6 kHz but it would reduce both hiss and speech in that part of the spectrum.

Alternatively you can use band stop filter for hum (power line) frequency and its harmonics - usually this is called NOTCH and means very narrow and high attenuation bandstop filter such notch has few Hz width so in theory should not severely impair signal and at the same remove interfering frequencies.

[timtape]

Technically if your noise is uncorrelated with signal then full separation (noise removal) is possible.

Yes but only if the noise is exactly the same signal in each channel but one channel's polarity is inverted. This is rare.

It help as i've wrote already for mono signal sampled as stereo and noise is uncorrelated between both channels.

Again it would need to be inverted between the stereo channels which is rare. It would have to have been recorded on a stereo recorder.

A mono recording on a tape can be captured with a stereo head, effectively splitting it into two halves. Any noise uncorrelated between the channels can be reduced by about 3db maximum by summing the channels. This is because while summing the correlated programme increases gain by 6db, summing the uncorrelated noise only increases the noise by 3db, hence the 3db noise reduction advantage. But the same advantage could have been obtained by just playing back the mono recording with a mono head in the first place.

A stereo linear audio head capture is useful because each of the narrower tracks of the stereo audio head is less sensitive to azimuth error than is a wider mono head. This is one of the reasons some pro S-VHS machines with true stereo linear audio heads reproduce recordings with better clarity. But only while the stereo separation is preserved. Summed to mono, the advantage disappears and the audio is just as muffled as if it had been captured with a mono audio head. There are good reasons to capture recordings in stereo but first we:

1. align azimuth to the actual recording while capturing in stereo but listening in summed mono. (This is rarely done partly because it usually involves opening up the deck to access the relevent adjusting screw. Most are advised not to attempt this without the relevent skill and knowledge)

2. Present the final product in summed mono.
3. If there are any remaining cancellations after summing channels to mono, we can use an "Azimuth" tool on the stereo capture such as in RX Advanced to dynamically realign the two channels before resumming to mono.

Yes that's been a proven technique for many decades but this is not the best example for it. Here the speech bandwidth extends up to about the same as the captured hiss. We could use a LPF below say 6 kHz but it would reduce both hiss and speech in that part of the spectrum.

Alternatively you can use band stop filter for hum (power line) frequency and its harmonics - usually this is called NOTCH and means very narrow and high attenuation bandstop filter such notch has few Hz width so in theory should not severely impair signal and at the same remove interfering frequencies.

Yes, in audio software such as Izotope RX the tool is called Dehum.

But Dehum is of no use in reducing the random broadband noise as heard in Wanderer's example. Random broadband noise remains a very difficult noise type to deal with.

[pandy]

Yes but only if the noise is exactly the same signal in each channel but one channel's polarity is inverted. This is rare.

Nope - uncorrelated noise is self canceling due lack of correlation - this is same principle as for amplifiers (for high end small signal amplifiers it is common to use many amplifiers in parallel - also quite common in ADC and DAC's) https://www.analog.com/en/resources/technical-articles/paralleling-amplifiers-improves...rformance.html

Again it would need to be inverted between the stereo channels which is rare. It would have to have been recorded on a stereo recorder.

A mono recording on a tape can be captured with a stereo head, effectively splitting it into two halves. Any noise uncorrelated between the channels can be reduced by about 3db maximum by summing the channels. This is because while summing the correlated programme increases gain by 6db, summing the uncorrelated noise only increases the noise by 3db, hence the 3db noise reduction advantage. But the same advantage could have been obtained by just playing back the mono recording with a mono head in the first place.

A stereo linear audio head capture is useful because each of the narrower tracks of the stereo audio head is less sensitive to azimuth error than is a wider mono head. This is one of the reasons some pro S-VHS machines with true stereo linear audio heads reproduce recordings with better clarity. But only while the stereo separation is preserved. Summed to mono, the advantage disappears and the audio is just as muffled as if it had been captured with a mono audio head. There are good reasons to capture recordings in stereo but first we:

1. align azimuth to the actual recording while capturing in stereo but listening in summed mono. (This is rarely done partly because it usually involves opening up the deck to access the relevent adjusting screw. Most are advised not to attempt this without the relevent skill and knowledge)

2. Present the final product in summed mono.
3. If there are any remaining cancellations after summing channels to mono, we can use an "Azimuth" tool on the stereo capture such as in RX Advanced to dynamically realign the two channels before resumming to mono.

Once again - you don't need to have inverted channels to be beneficial by using many channels - even your stereo ADC can sample mono signal and after combining both channels in digital domain (mathematically lossless) you are still beneficial and you improve your SNR for uncorrelated noise (so it will be mostly noise produced by your ADC).

Yes that's been a proven technique for many decades but this is not the best example for it. Here the speech bandwidth extends up to about the same as the captured hiss. We could use a LPF below say 6 kHz but it would reduce both hiss and speech in that part of the spectrum.

Alternatively you can use band stop filter for hum (power line) frequency and its harmonics - usually this is called NOTCH and means very narrow and high attenuation bandstop filter such notch has few Hz width so in theory should not severely impair signal and at the same remove interfering frequencies.

Yes, in audio software such as Izotope RX the tool is called Dehum.

But Dehum is of no use in reducing the random broadband noise as heard in Wanderer's example. Random broadband noise remains a very difficult noise type to deal with.

Oh for sure - narrow notch filter is tailored for removing almost statically located interferer from signal such as power line frequency with harmonically related.

For dynamic hiss you need to have some form of adaptivity - you can give a chance to ffmpeg neural trained noise removal filter https://onelinerhub.com/ffmpeg/how-to-reduce-background-audio-noise-using-arnndn - perhaps it can help

[timtape]

Nope - uncorrelated noise is self canceling due lack of correlation - this is same principle as for amplifiers (for high end small signal amplifiers it is common to use many amplifiers in parallel - also quite common in ADC and DAC's) https://www.analog.com/en/resources/technical-articles/paralleling-amplifiers-improves...rformance.html...


Once again - you don't need to have inverted channels to be beneficial by using many channels - even your stereo ADC can sample mono signal and after combining both channels in digital domain (mathematically lossless) you are still beneficial and you improve your SNR for uncorrelated noise (so it will be mostly noise produced by your ADC).

Yes of course. I once constructed a super low noise preamplifier for a moving coil phono cartridge using an LM394 IC which operated on that principle. It worked well. Many VCR linear audio track preamps could benefit by upgrading to a small signal preamp based on that principle. It can usefully reduce the original preamp's noise contribution especially in a VHS LP or EP recording where the audio signal is very weak. But of course it will not reduce the noise in the recording itself, unfortunately.

For dynamic hiss you need to have some form of adaptivity - you can give a chance to ffmpeg neural trained noise removal filter https://onelinerhub.com/ffmpeg/how-to-reduce-background-audio-noise-using-arnndn - perhaps it can help

There are limits to what noise can be reduced without also damaging the wanted audio. Manufacturers of noise reducing software typically demonstrate with easy samples than make their product look good.

I have made up a severe test recording where the noise is so strong that the wanted voice underneath it is unintelligible.
Anybody is welcome to try the latest software on the test sample to make the voice intelligible and upload their result. See the link below.

https://files.videohelp.com/u/310643/mixed%20sample.wav

[davexnet]

I have made up a severe test recording where the noise is so strong that the wanted voice underneath it is unintelligible.
Anybody is welcome to try the latest software on the test sample to make the voice intelligible and upload their result. See the link below.

I suspect the forensic audio tools, even those generally available such as Diamond Cut Audio could do it.
Whether there is anybody here proficient in its use is another question

[timtape]

I have made up a severe test recording where the noise is so strong that the wanted voice underneath it is unintelligible.
Anybody is welcome to try the latest software on the test sample to make the voice intelligible and upload their result. See the link below.

I suspect the forensic audio tools, even those generally available such as Diamond Cut Audio could do it.
Whether there is anybody here proficient in its use is another question

It's hard to prove that it cant be done. But it only takes one competent person with one capable tool to successfully prove that it can!

[pandy]

I have made up a severe test recording where the noise is so strong that the wanted voice underneath it is unintelligible.
Anybody is welcome to try the latest software on the test sample to make the voice intelligible and upload their result. See the link below.

This is recording prepared intentionally but if you prepare recording where there is few seconds without signal (like mute) with only noise then this is way more possible. - good habit is to left few second silence at beginning of recording so you can remove own equipment noise.

[pcspeak]

@Wanderer.
Trying to keep it as straight forward as possible.
DePopper64 (30 day trial)
Default settings - run a second time against the first output.

Then the second output as input:
Audacity
Effect/Noise Reduction
Default settings - run 3 times.
Noise profiles taken from 3 different places.

[davexnet]

That's brilliant. Didn't even need forensic software - just a vinyl cleanup program and Audacity

[timtape]

That's brilliant. Didn't even need forensic software - just a vinyl cleanup program and Audacity

In Wanderer's original example there is significant background noise but the speech is fully intelligible. But this "cleanup" makes intelligibility worse.

I dont think you understand that forensic audio is all about improving or at least preserving speech intelligibility to be used for evidence in court cases. Judged on that score, this processing has failed because it has made intelligibility worse. For example listen when the interviewer says "pushing up against this..." (about 0:14 seconds) In the original the word "against" is clear. In the processed version the word "against" is now harder to understand.

This interview is obviously not intended for evidence a court case but surely the listeners interested in the interview want to understand what is being said!

And here we only hear a 16 second clip of the interview. In the full interview we dont know what other words might have been also distorted or lost due to this sort of "cleanup". I dont mean to single out any one person. This sort of thing is unfortunately quite common these days.

[davexnet]

No I understand. I guess it would be up to the court and legal representatives to decide if what was obtained
was admissible or not.
Perhaps I misunderstood this whole thing:
What point are you trying to make by creating this test? That the original quality cannot be derived from your file?
Some criticism of noise reduction technology in general? Something else?