Subtle hissing after noise reduction?

[Rapunzel]

Forgive me if this is a stupid question, I've just switched to Adobe Audition so I'm still figuring things out. I use the noise reduction tool and it works great on the dead air, but when I'm speaking there's a subtle hissing noise. I've attached a sample so you can see what I'm talking about. Am I just being a perfectionist, or is there something to be done to make my speaking clearer? You can especially hear it in the "Super Rabbit Boy" section of the sample.

I'm recording with a ShureSM7B, mono, 44.1hz, 16bit. Editing on a MacBook with Adobe Audition.

Thank you so much! I'm on my own learning how to edit so I really appreciate any advice you can give me!

Sample
https://soundcloud.com/user-494478489/t ... Z8NQrJBhbR

[Tim Gillett]

Hi Rapunzel and welcome.

I worked in audio book production for many years.

You've described well one of the side effects of many noise reductions tools: a hissy/furry sound when we speak. When the noise reduction tool opens up to let your voice through, the hiss of the "dead air" can slip through as well. I'm guessing that's what you're hearing.

Noise reduction can be fine if we tend to speak all at the same level, but if we use dynamics for dramatic effect and speak quietly at times, another problem can arise: the NR assumes that if it's quiet it's unwanted and it can push your quiet voice even quieter, unnaturally so. So be careful with NR especially if you tend to speak with dramatic range.

The Sm7b is known for its fairly weak signal which tends to exaggerate hiss. You could try adding a "Cloudlifter" type booster after the mic which may help reduce the hiss in the first place. Or perhaps use a different mic with a stronger output.

Hope this helps,

Tim.

[The Elf]

I'd say reduce as much noise as you can at source, then accept what is left. Use of noise reduction should really be an absolute last resort - personally I wouldn't use it at all.

[Hugh Robjohns]

I use the noise reduction tool and it works great on the dead air, but when I'm speaking there's a subtle hissing noise. I've attached a sample so you can see what I'm talking about. Am I just being a perfectionist...

Yes!

That's a very, very dynamic voice presentation you have there! I had to turn my monitoring up to silly levels before I became aware of the hiss, by which time your transients were positively painful.

As Tim has said, some noise reduction systems will introduce the side-effect of a noise modulation around the voice, and that can become more obvious with such a staccato and dynamic voice as yours.

I don't know what you've used or how you've applied your noise reduction strategy, or the problem you're actually trying to improve... but in general, a low level constant hiss or noise floor is generally considered more acceptable and less attention-grabbing than a varying, modulated noise floor. And as the Elf says, noise-reduction should really be a last resort... it's always much better to reduce unwanted noise at source.

But if it's the remaining high-frequency hiss that's troubling you, it seems to me that there is plenty of high-frequency energy in your recording anyway, so you could afford to apply a high-frequency roll-off EQ to tame a little of the voice brightness while also reducing the audibility of the HF hiss.

[Wonks]

And as nobody else has mentioned it, you should really be working in 24 bit when recording, unless there's a technical reason you can't.

If the final product output needs to be 16-bit, then convert at the end after editing and processing.

[The Elf]

And as nobody else has mentioned it, you should really be working in 24 bit when recording, unless there's a technical reason you can't.

Well spotted!

[Rapunzel]

You could try adding a "Cloudlifter" type booster after the mic which may help reduce the hiss in the first place. Or perhaps use a different mic with a stronger output.

Hope this helps,

Tim.

Thank you! I am using a cloudlifter, I forgot to mention that. Sorry, it was like 4 in the morning when I wrote that post haha

And as nobody else has mentioned it, you should really be working in 24 bit when recording, unless there's a technical reason you can't.

If the final product output needs to be 16-bit, then convert at the end after editing and processing.

Oh really?? I had no idea! ACX said that things had to be 16 bit so I assumed that meant I had to record in that. I didn’t know I could change it later on. Thank you so much! I’m still learning how to do all this with just the help of Professor Google lol

I use the noise reduction tool and it works great on the dead air, but when I'm speaking there's a subtle hissing noise. I've attached a sample so you can see what I'm talking about. Am I just being a perfectionist...

Yes!

That's a very, very dynamic voice presentation you have there! I had to turn my monitoring up to silly levels before I became aware of the hiss, by which time your transients were positively painful.

As Tim has said, some noise reduction systems will introduce the side-effect of a noise modulation around the voice, and that can become more obvious with such a staccato and dynamic voice as yours.

I don't know what you've used or how you've applied your noise reduction strategy, or the problem you're actually trying to improve... but in general, a low level constant hiss or noise floor is generally considered more acceptable and less attention-grabbing than a varying, modulated noise floor. And as the Elf says, noise-reduction should really be a last resort... it's always much better to reduce unwanted noise at source.

But if it's the remaining high-frequency hiss that's troubling you, it seems to me that there is plenty of high-frequency energy in your recording anyway, so you could afford to apply a high-frequency roll-off EQ to tame a little of the voice brightness while also reducing the audibility of the HF hiss.

Thank you so much! I’m glad to know it’s not glaringly obvious—I always focus on the mistakes so it’s all I can hear when I’m editing. I’ll try the high frequency roll off and see if that helps!

[James Perrett]

One other thing with noise reduction, especially the one in Audition, is to not do too much at once. I use the older Audition 3 and I rarely do more than 12dB of reduction in one pass and I never set the reduction fader to more than 50%. If you need more than 12dB of noise reduction use two or three passes.

[Hugh Robjohns]

Oh really?? I had no idea! ACX said that things had to be 16 bit so I assumed that meant I had to record in that. I didn’t know I could change it later on.

ACX want the release material to have a 16-bit wordlength because it gives more consistent results in conversion to the MP3 format for their distribution arrangements.

However, most professionals prefer to record and mix/process with longer wordlengths because it affords a much larger dynamic range. The practical outcome is that you can record with more headroom -- so you won't get caught out with occasional overloads -- while still keeping the signal well above the system's noise floor.

I don't know what interface you're using, but it will almost certainly be passing 24-bit audio data into Adobe Audition, and Audition itself works internally with 32bit (floating point) calculations.

So configure your next project to work with either '24 bit' or '32 float'. And when you record, adjust the input levels so that the highest peaks on your voice go no higher than -6dBFS. I usually aim for -10dBFS, knowing that a performance is always louder than a level check!

Record and edit as usual, but once you've finished you may wish to remove the now redundant headroom margin (since you now know exactly how loud the highest peaks are). The easiest way is to select everything and 'normalise' to a level you set. ACX require peaks to be at -3dBFS. The normalise function is under Effects > Amplify & Compression.

Some people prefer to perform the normalise function straight after recording, as it raises the overall level and makes it easier to hear (and see) what's going on. However, I prefer to maintain the headroom margin until I've finished all the signal processing and editing... swings and roundabouts...

Finally, with the editing/mixing complete, and the file normalised to comply with the ACX requirements, you can archive the 24-bit file just in case you need to work on it again inthe future, and then export the file for the ACX submission.

This is when you convert to 16 bits, and there is an option in the 'sample type' box to select the wordlength. Change to 16 bit and make sure 'Dithering' is enabled with 'Triangular' as the option and 'none' in the 'Noise Shaping' box. Once you have the 16 bit file, you can then convert to MP3 at 192kb/s CBR as they request.

Hope that helps.

[Wonks]

'Triangular' as the option and 'none' in the 'Noise Shaping' box. Once you have the 16 bit file, you can then convert to MP3 at 192kHz CBR as they request.

Or even 192 kbit/s CBR.

[Hugh Robjohns]

Oops, sorry. fingers in autocomplete mode with brain disengaged! I've corrected my original!

[Tim Gillett]

Rapulzel, it would be good if in addition to the denoised file, you could upload the unprocessed original as well. We could then offer comments on the hiss you mentioned and what might generally be the best options going forward.

Tim.

[Tim Gillett]

I'd say reduce as much noise as you can at source, then accept what is left. Use of noise reduction should really be an absolute last resort - personally I wouldn't use it at all.

Great to hear someone say this. Working with vintage recordings - which unlike here obviously cant be remade - I can feel I've made progress if I can reduce noise by even one db here, and perhaps another db there, by careful cleaning of the original media, maintenance, selection and alignment of the read head or stylus to the media, etc.

Then I read of others who think nothing of achieving 12 or even 24 db of broadband noise reduction, and just at the press of a button or two! It's a different world. It can seem like we're from different planets, talking different languages, but perhaps more importantly hearing different things.

On this audiobook example here I couldn't hear hissing artifacts at all but that's probably due to ageing ears which these days cant perceive anything above about 10 kHz.

[innerchord]

The example does sound a little over-processed, and possibly a bit too sibilant at source. I assume it's not a finished section due to the lack of dynamic control in the audio.

The SM7B may not be the best choice for this voice, but it's perfectly usable and there's no reason you shouldn't get great results with next to no noise. If you're a stickler for really quiet backgrounds, you could use, say, 8-10dB of NR. Make sure your noise samples (noise profiles) accurately reflect the noise at the time of recording.

You are using a good (quiet) audio interface, I presume?

Lastly, I wouldn't use any filtering for this voice. A good de-esser would be my instant choice if you're still worried about the brightness (by I would try to fix it at the source first...).

[Mike Stranks]

I'd say reduce as much noise as you can at source, then accept what is left. Use of noise reduction should really be an absolute last resort - personally I wouldn't use it at all.

Great to hear someone say this. Working with vintage recordings - which unlike here obviously cant be remade - I can feel I've made progress if I can reduce noise by even one db here, and perhaps another db there, by careful cleaning of the original media, maintenance, selection and alignment of the read head or stylus to the media, etc.

Then I read of others who think nothing of achieving 12 or even 24 db of broadband noise reduction, and just at the press of a button or two! It's a different world. It can seem like we're from different planets, talking different languages, but perhaps more importantly hearing different things.

On this audiobook example here I couldn't hear hissing artifacts at all but that's probably due to ageing ears which these days cant perceive anything above about 10 kHz.

Tim:

You've taken The Elf's comment about NR in this specific case and extrapolated it into another discourse on NR in general.

Will you please accept that for those of us who have to deal with received audio of sometimes very poor quality, the use of NR is essential?

Please also stop talking about NR in such a broad and ill-defined way. NR is much more than taking an analysis of the noise and then applying a polarity invert to the recording. There are numerous tools available which avoid such relatively crude processing. Used with care these specific, targetted, tools achieve very good results indeed.

But we've been here before and you've been encouraged to try some of these specific tools for yourself. By failing to reply to questions about your use of them I can only infer that you have declined to do so.

But yet you still expound your negative views which would have more credibility if you were able to say that, "On the sample you supplied I used the 'x' NR tool and noticed that..."

Abstract theorising is fine in academic circles; some of us live in a production world where professional NR tools are an essential part of our working practices.

But I agree with The Elf, probably not in this case...

[Hugh Robjohns]

With the greatest respect to all contributors, we've been around this loop so many times before, but apparently no change in viewpoints or understanding. So I think I'll lock this thread to avoid further bitter but fruitless exchanges...