Identifying offending noise by using spectral frequency display

[Kathy Smith]

Hi,

I have a recording where there is a bang happening during the interview. I'm trying to remove it using Izotope RX but I'm having a hard time isolating the offending sound. Here is a screenshot of the spectral frequency display and I'm attaching that section of the audio that has that bang sound. Can someone tell me where on the spectral frequency display that sound is? Do you think it's possible to remove this sound? I have this banging happening in 3 sections of the recording. It would be nice to be able to remove it

THANK YOU!

[Alexey Lukin]

The bang overlaps speech on a spectrogram here, so it's not easy to get rid of it. You'll need to use Deconstruct instead of Spectral Repair.

[Jim Andrada]

Hi Kathy

Do you have a bit longer clip so we could get a better idea of the pre and post "bang" content - maybe 10 - 15 seconds or so?

[Kathy Smith]

Originally Posted by Jim Andrada

Hi Kathy

Do you have a bit longer clip so we could get a better idea of the pre and post "bang" content - maybe 10 - 15 seconds or so?

Hi Jim,

Here is the file with more content. Is this enough?

[Kathy Smith]

Originally Posted by Alexey Lukin

The bang overlaps speech on a spectrogram here, so it's not easy to get rid of it. You'll need to use Deconstruct instead of Spectral Repair.

Thanks Alexey, I don't seem to have Deconstruct in my version of RX, is it part of the Advanced version?

[Alexey Lukin]

Yes, it's Advanced-only.

[Kathy Smith]

Originally Posted by Alexey Lukin

The bang overlaps speech on a spectrogram here, so it's not easy to get rid of it. You'll need to use Deconstruct instead of Spectral Repair.

Originally Posted by Alexey Lukin

Yes, it's Advanced-only.

:(

Thank you for your help though. May I ask how do you actually know that that's where the bang is on the spectrogram?

[Alexey Lukin]

I have listened to a few selections in your file. My first suspicion was the vertical line before my selection, but it turned out to be letter 'k'. Finally I have arrived at the correct selection.

When you have experience with audio, you can estimate the frequency content of any sound. In your case, it's a noise burst at lows and mids. And it has a considerable decay, due to reverberation.

[Kathy Smith]

Originally Posted by Alexey Lukin

I have can listened to a few selections in your file. My first suspicion was the vertical line before my selection, but it turned out to be a letter 'k'. Finally I have arrived at the correct selection.

When you have experience with audio, you can estimate the frequency content of any sound. In your case, it's a noise burst and lows and mids. And it has a considerable decay, due to reverberation.

Thank you. I also thought it was the vertical line but when I isolated it I learned it wasn't it. Thank you

[Kathy Smith]

I downloaded the trial of the Advanced version of RX. I won't be able to save it anyway but I thought I would try to see what I could possible achieve but I'm not able to get a good result either. Makes me think it's not possible to remove this. I wish I could at least lessen it. Anyone has any other ideas?

[Jim Michael]

Can you select the speech harmonics in Izotope? That's the procedure in SpectralLayers.

[Battle Vaughan]

A quick and dirty try with noise reduction in (freeware) Audacity makes some improvement, imperfect but less noticeable. The sound seems to mostly fall between "the" and "com" and between "com" and "part" and by taking small slices of the intervening space and doing a 6 to 12 db reduction on those samples (three) you can get what might pass as a momentary blip that most people would disregard, rather than a burst of hiss that does stand out. Then, applying the sample to the "com" phrase removes some of the underlying hiss without destroying the vocal.
Also reducing the pause after "-mentalization" takes care of the rest. You do it in pieces, a little at a time....

Fortunately this sound falls largely in a pause in the narration, making the reduction samples possible. Now, if the sound is the same in the other places, it might be possible to use the samples from the first situation and apply them (one at a time, obviously) to the others, hopefully then reducing the sound and not affecting the speech too much. (If the second and third sounds overlap the speech, you'd not want to sample there because the speech would be part of the reduction sample.)

[Kathy Smith]

Originally Posted by Jim Michael

Can you select the speech harmonics in Izotope? That's the procedure in SpectralLayers.

I don't think so. At least I don't know how

[Kathy Smith]

Originally Posted by Jim Michael

Can you select the speech harmonics in Izotope? That's the procedure in SpectralLayers.

Originally Posted by Battle Vaughan

A quick and dirty try with noise reduction in (freeware) Audacity makes some improvement, imperfect but less noticeable. The sound seems to mostly fall between "the" and "com" and between "com" and "part" and by taking small slices of the intervening space and doing a 6 to 12 db reduction on those samples (three) you can get what might pass as a momentary blip that most people would disregard, rather than a burst of hiss that does stand out. Then, applying the sample to the "com" phrase removes some of the underlying hiss without destroying the vocal.
Also reducing the pause after "-mentalization" takes care of the rest. You do it in pieces, a little at a time....

Fortunately this sound falls largely in a pause in the narration, making the reduction samples possible. Now, if the sound is the same in the other places, it might be possible to use the samples from the first situation and apply them (one at a time, obviously) to the others, hopefully then reducing the sound and not affecting the speech too much. (If the second and third sounds overlap the speech, you'd not want to sample there because the speech would be part of the reduction sample.)

Thanks, I did the same thing you described but in Izotope Rx and managed to make it less noticeable.

[Jim Andrada]

Here's what I came up with just for the record