Salvaging a under recorded source

[twotoedsloth]

Hello,

I'm recording a classical vocal performance, but unfortunately we didn't get the chance for a sound check. As such, I'm running very low gain to avoid clipping.

I'm going to need to do some noise reduction to control the hiss, should I do the noise reduction before normalizing or after? What is the best practice?

Thanks,

Peter

[Drew Stephenson]

I would always do noise reduction first, give it the most unadulterated signal to work on and it will generally work at its best.

[twotoedsloth]

Thanks for your advice. It does seem solid.

The only problem that I've encountered in the past is that the piano tails fade out to the noise floor, so with a normalized sound print there's more to work with. I can't check this now, as the concert is still going on, but I'll do both and compare the results.

Again, thanks for your reply,

Peter

[James Perrett]

Mathematically it should make no difference to a good noise reduction algorithm whether you use a normalised signal or the original level. However, your mention of tails disappearing makes me wonder whether you are using a crude noise reduction algorithm or if you are trying to do too much noise reduction in a single pass.

Noise reduction is the one area where I've found RX hard to beat.

[Hugh Robjohns]

I'm recording a classical vocal performance, but unfortunately we didn't get the chance for a sound check. As such, I'm running very low gain to avoid clipping.

What's the highest recorded peak level?

It would be unusual to have a noise problem due to the (professional) gear. It's normally ambient room noise that dominates.

I'm going to need to do some noise reduction to control the hiss, should I do the noise reduction before normalizing or after?

Shouldn't make any real difference... but I'd normalise to -6dBFS and then process, personally. And two or three passes with 4-6dB noise reduction each generally sounds better than one pass with 12-15dB noise reduction.

Peter
[/quote]

[twotoedsloth]

Hello,

Thank you for your help with this.

Now that the concert has ended, I looked at the file, and the peak level is -19db. I usually normalize to -3db. When I was recording the level looked much lower than -19, which is why I wrote the original post. Now normalized, there really isn't that much noise, so I'm not going to worry about it.

If someone complains, I do have options, I have licenses for the ancient DINR, the equally antiquated Sound Forge NR, and the one I've been using for many years now: Izotope RX. I will try your method of multiple passes next time!

Many thanks,

Peter

[Hugh Robjohns]

Now that the concert has ended, I looked at the file, and the peak level is -19db.

I wouldn't call that 'under-recorded' at all. Ideally, a rehearsal would have allowed you to get maybe 10dB louder, but that peak level is certainly not an issue in a 24bit recording.

I have live choir recordings with solos/duets barely peaking -30dBFS, but they're still well above the electronic noise.

Now normalized, there really isn't that much noise, so I'm not going to worry about it.

I'd agree!

Izotope RX. I will try your method of multiple passes next time!

It really does help if you need heavy NR.