Restoration and denoising trade-off

[PBM]

Originally Posted by Lzon

Hi everyone,
I'm working on restoring a piece of classical music. I have been successful in removing clicks, crackle and hum, but I am hitting a bit of a brickwall in removing the broadband noise. It is no problem during the louder sections of music, but during the quieter passages the noise is as loud as the desired material.
I'm beginning to feel that there will be some inevitable compromise between artifacts and residual hiss. Currently I'm thinking of noise-reducing as far as I can during these sections before artifacts are introduced and accepting a certain amount of hiss.

While I understand that digital artifacts are undesirable, my aim is to create a version as close to the original intended sound, and I feel that any evidence of the recording medium is unacceptable.

Is there a way of avoiding this? If not, does anyone have any advice on how to go with this decision?
Thanks everyone!

I've done a lot of this kind of work - using both CEDAR and RX2 - and the fact is that compromise is unavoidable. On the brighter side of life: if this is an older recording then people tend to expect, and accept, a bit of hiss; and it is amazing how quickly the ear accommodates it (evidence of post-hoc tampering is much more objecitonable).

Some of my favourite chamber music (the Hollywood and Italiano string quartets) has fairly obvious hiss but that's never what I'm listening to.

Cheers,

Eric

[Zownd]

Originally Posted by Lzon

Hi everyone,
I'm working on restoring a piece of classical music. I have been successful in removing clicks, crackle and hum, but I am hitting a bit of a brickwall in removing the broadband noise. It is no problem during the louder sections of music, but during the quieter passages the noise is as loud as the desired material.
I'm beginning to feel that there will be some inevitable compromise between artifacts and residual hiss. Currently I'm thinking of noise-reducing as far as I can during these sections before artifacts are introduced and accepting a certain amount of hiss.

While I understand that digital artifacts are undesirable, my aim is to create a version as close to the original intended sound, and I feel that any evidence of the recording medium is unacceptable.

Is there a way of avoiding this? If not, does anyone have any advice on how to go with this decision?
Thanks everyone!

I've been working some live recordings and although I don't have the cash for cedar I came a long way.

With waves it's very easy to get rid of hummm and if you have a sample profile of the noise you can get very far with Z-noise or X-noise. If money allows it the new Sonnox plugins are amazing.... all of them have demo's. You can try Izotope, but that is kids stuff. A lot of artifacts or remaining noise. Sonic's "no noise" is supposed to be very good, but I never seem to get good results.

Anyway you should target the part of the spectrum that reveals most of the noise. Then make a profile of that. Commonly I use a hipass filter to get to the part where noise is most evident and make a profile of that... this can truly do miracles.

If you feel uncertain, drop me a PM and maybe we can work something out

[Riccardo]

Sonic is very good, Cedar too.
I have never used Sonnox so i can't comment.
Although sometimes a compromise has to be struck on some material regardless of what NR products are used I did not find RX or RX advanced kids stuff as Zownd (see above post) but on the contrary quite good.
I guess it also depends on the source material.

[lowland]

I have a friend who is a restoration specialist, and can achieve near-miraculous results with extremely challenging material. RX is very much part of his toolkit, although he also has just about everything else including the Weiss box -

Weiss :: DNA1

He says you sometimes don't know what tool will work best until you try, effectiveness has only a loose connection with money spent and it's not unusual for the thing that gets the best result to do so in a counter-intuitive way. Sounds like fun to me!

[PBM]

Originally Posted by Zownd

You can try Izotope, but that is kids stuff.

Utter rubbish: we very successfully restored a whole series of classic opera recordings from the Sydney Opera House and Teatro Real last year using RX2. It is a superb tool.

[polybonk]

When noise reduction with the usual suspects doesn't work. Sometimes a really nice eq and filter in parallel works wonders.

[huejahfink]

Originally Posted by PBM

Utter rubbish: we very successfully restored a whole series of classic opera recordings from the Sydney Opera House and Teatro Real last year using RX2. It is a superb tool.

I completely agree with PBM. Izotope is far from kids stuff.
I don't want to be overly presumptuous but that comment makes me think that you (Zownd) have not fully explored the myriad of options available in RX2...
All versions of RX are very capable, but the advanced version in particular is highly tweakable, allowing you to audition various types of artefact suppression, gentle harmonic enhancement to compensate for slight loss in highs, and various weightings between FFT fingerprinting and wider gating.

I've been using RX for a while, always been happy with the results. I have always been under the impression that the RX line was Izotope's most widely respected software release amongst professional users, and while possibly not as highly regarded as the offerings from Cedar and Algorithmix in some circles, certainly considered to be the most able in its price class.
Kids stuff? 3000% disagree.

[Lzon]

Thanks for your responses! I'm going to give the EQ before noise learning a go and also experiment with EQ in z-noise.
If that fails I'll try the sonnox demo and some EQ!
I have used RX2 for a lot of film work and loved it, but in this case the waves plug-ins have been more effective. As lowland hinted, I feel like I've got to try everything and won't know until I've explored every option.. Will post back with developments..

[Alexey Lukin]

Originally Posted by Lzon

I'm beginning to feel that there will be some inevitable compromise between artifacts and residual hiss. Currently I'm thinking of noise-reducing as far as I can during these sections before artifacts are introduced and accepting a certain amount of hiss.

Posting a WAV or AIFF sample would be helpful here.

Depending on your recording you may need a combination of different noise reduction means: Denoiser, Deconstruct, Azimuth Corrector + Vocal Extractor... They all produce slightly different denoising results, and can be averaged together in hard cases.

[Zownd]

Originally Posted by PBM

Utter rubbish: we very successfully restored a whole series of classic opera recordings from the Sydney Opera House and Teatro Real last year using RX2. It is a superb tool.

Well I guess I will have to give it another go then
My findings with all the tools available was that RX performed very poorly,.. sorry.

But I must confess that material can change everything.
Although Sonnox has been my tool to go for and I have seldom gone for an other option, simply because it was not necessary.

[nevets@milwpc.co]

Voxengo redunoise is very good. I've done 10 CD's with this. The only reason it's not more popular is because the author doesn't tell you the very first thing to do to use it. You can just ask on the forums. It's very simple after that.
If you're dealing with tapes here's a couple tips;

Redunoise works best on 44.1/16 bit files.
Try doing an eg boost/reduction before and after using noise reduction. There's no telling when this can be very effective. Try equing about -10db on the low end and +10db on the high end-do the noise reduction and then eq the opposite, +10db low and -10db high. Somtimes it works, sometimes not.
Don't try to get rid of all the noise. Shoot for about half and see how it goes.
I tried everything as far as Vitalizers, harmonic enhancers, etc... to restore some of the sound but it's been my experience that nothing works for classical. Trying to restore the sound of washed out cymbals on a rock or pop track is much different than trying to restore the overtones on classical recordings. If they're gone they're gone.

Just my 2 cents.

[Hane]

Originally Posted by Alexey Lukin

Posting a WAV or AIFF sample would be helpful here.

Depending on your recording you may need a combination of different noise reduction means: Denoiser, Deconstruct, Azimuth Corrector + Vocal Extractor... They all produce slightly different denoising results, and can be averaged together in hard cases.

This might be slightly off topic but Alexey, is it possible in future updates to RX to de-noise music using the finger print method using the spectrogram editor?

Say for example you want to de-noise a file by 10db but, you don't want to use the time v amplitude editor because the noise you want to remove is easier to see in the spectrogram editor.

Another problem i have is removing hum from music with prominent bass lines.
The de-hummer does a good job in removing hum but, it's also removing the bass notes which falls on the same frequency as the 50Hz/60Hz hum.

To get round this problem; I'm manually cutting away the hum in the spectrogram and when the bass note hits 50Hz or 60Hz, I'm relying on the masking effect to hide the hum. Is it possible in future versions to have a de-hummer that detects the constant hum tone that reduces it by the amount of decibels you set and, when it detects a sudden change in amplitude caused by for example a bass guitar, it reduces the hum reduction to 0db.

[huejahfink]

Originally Posted by Hane

This might be slightly off topic but Alexey, is it possible in future updates to RX to de-noise music using the finger print method using the spectrogram editor?

Say for example you want to de-noise a file by 10db but, you don't want to use the time v amplitude editor because the noise you want to remove is easier to see in the spectrogram editor.

I wasn't sure what you meant at first - but now I presume you mean that you wish you could train the denoiser from a time+freq selection on the spectral display, rather than having to play through the all the frequencies in a time selection?
Good idea, could just have a dedicated button / keyboard shortcut to train the denoiser once you have made your selection.
As a work around, I have often taken a time / freq selection of the audio I wish to train with, copied that across to a section of silence at the end of the file so that the denoiser can be trained without the intrusion of the other frequencies.
Not ideal, but doesn't take that long really.

Originally Posted by Hane

Another problem i have is removing hum from music with prominent bass lines.
The de-hummer does a good job in removing hum but, it's also removing the bass notes which falls on the same frequency as the 50Hz/60Hz hum.

To get round this problem; I'm manually cutting away the hum in the spectrogram and when the bass note hits 50Hz or 60Hz, I'm relying on the masking effect to hide the hum. Is it possible in future versions to have a de-hummer that detects the constant hum tone and reduces it by the amount of decibels you set and when it detects a sudden change in amplitude caused by for example a bass guitar, it reduces the hum reduction to 0db.

Yes I have also thought this. A dynamically sensitive de-hum could be useful. However, have you considered trying the denoiser in tonal mode only (ie, switch the broadband noise reduction off) trained using 'just' a section of hum? Probably not the same kind of results, but worth a shot, since the denoiser IS threshold sensitive.

[Bruno Republic]

One trick which works with vinyl recordings is to run it through a mid-side decoder, and process the channels separately (also useful for isolating certain clicks, sibilance from tracing distortion, and low-frequency thumps). There will be more "black noise" in the side than the mid, and depending on how the record was mastered, you may be able to be much more aggressive with NR on the side channel, particularly in the lower frequencies, while doing less damage.

[Alexey Lukin]

Originally Posted by Hane

Another problem i have is removing hum from music with prominent bass lines.

If you have a section of pure hum, try Denoiser "C" instead of Hum Remover. In many cases, it produces more accurate results, because it would not cut the bass that overlaps hum. I can recommend some particular settings if you post an example.

We've heard the request to train denoiser from arbitrary spectrogram selections before. We do consider it for future upgrades.

[Hane]

Originally Posted by huejahfink

I wasn't sure what you meant at first - but now I presume you mean that you wish you could train the denoiser from a time+freq selection on the spectral display, rather than having to play through the all the frequencies in a time selection?

That's what I'm aiming for.
The ability to take a finger print of noise within a specific frequency range that I highlighted in the spectrogram and RX reduces the noise in the entire file in the same manor as it would in the amplitude v time editor.

Originally Posted by huejahfink

Yes I have also thought this. A dynamically sensitive de-hum could be useful. However, have you considered trying the denoiser in tonal mode only (ie, switch the broadband noise reduction off) trained using 'just' a section of hum? Probably not the same kind of results, but worth a shot, since the denoiser IS threshold sensitive.

I'll give that method a try. I assumed the de-noiser in (tonal mode only) would treat the bass notes as noise.

[Hane]

Originally Posted by Alexey Lukin

If you have a section of pure hum, try Denoiser "C" instead of Hum Remover. In many cases, it produces more accurate results, because it would not cut the bass that overlaps hum. I can recommend some particular settings if you post an example.

We've heard the request to train denoiser from arbitrary spectrogram selections before. We do consider it for future upgrades.

I'll give your approach a try as well, thanks.

[huejahfink]

I think myself and Alexey are basically talking about the same method, only he suggested to use type C (which I always use anyway)

[16/44.1]

Originally Posted by polybonk

When noise reduction with the usual suspects doesn't work. Sometimes a really nice eq and filter in parallel works wonders.

Better that than noise reduction