How to remove "fry" - vocal raspiness?

Hello, I’ve gone round in circles trying to remove a kind of rumbly, raspy sound in a voiceover track. It’s the main reason why I purchased SpectraLayers.

It seems to occur intermittently and although I’ve tried to select various areas in the audio, I can’t identify where it’s coming from (because once I do then at least I could remove or ameliorate it).

Also, I don’t know how much overlaps with the actual “voice” sound - I don’t want to remove or affect the voice itself - just remove the “rumble” that may be coming from somewhere in the throat or perhaps a vibration of the bones in the mouth or something!

Any idea where to start?

Listen especially to the words: throughouttimeswhen I was metdoubtI faced four (decades ago).

Try this awesome thread!

haha, I started that thread here a good two years ago. I haven’t found a good solution yet. The main problem is identifying which part of the audio signal is the crackly or raspy bits.

That’s why I thought it better to start a new thread. Once you can accurately identify the “noise” elements, then one can run Spectral Denoise on Izotope RX Advanced or Find Similar on Spectralayers.

What I like about Spectralayers so far is that you can select parts of the signal with the paintbrush tool. But for the life of me, I can’t find the noise (or at least separate it out from the voice signal).

Maybe there’s a different way to visualize the audio. That would probably nail it!

Try this (a bit tedious selecting process though…):

  1. Concentrate on the frequencies where the raspiness is the most pronounced, that is up to 2,2 k. Select this range with the Frequency Range Selection tool.
  2. Copy to new layer.
  3. On this new layer: use the Harmonics Selection tool to select the spoken words (this could be tedious…); use Thickness 2 pixels, Master Rank 1 and Count 35. Select with the mouse pointer on the lowest “spoken words bands”.
  4. Erase the selected areas (Backspace multiple times). The resulting material will be the warbles/rasp.
  5. Phase invert this Layer and play it together with the original layer; Merge the layers when satisfied. The result should be significantly less raspy.

Much better!

Seems like the tools aren’t good enough for complete transparency yet … sound processed to me.

I should add I’ve used a simple LPF, e.g., down to as low a 5-7K in a male singer, to get rid of a singer’s vocal fry. I guess it would depend on the musical setting, but in the right circumstances it can sound quite transparent.

Nasel passages are causing the added noise, IMO. I can’t imagine how something so ingrained in the delivery could be separated from the final product.

I have had success by removing the “wolf” frequencies between the lower 5 partials. When you look at your audio, you will see the pure fundamental frequency and another frequency between it and the 2nd partial (the actual location of the wolf frequencies will vary depending on the vocalist, but they will be in between the pure partials.

I perform the work using a group, layers, phase inversion, and composite view.

  1. Create a new group
  2. Place the source audio in the group
  3. Add another layer for the pure vocal frequencies (SPEECH)
  4. Add another layer for the noise partials causing the “fry.” (NOISE)… (Be aware that polarity in v9 is messed up. If you have v8, use it.)
  5. Enable composite view

Locate an area of vocal fry.
Use the frequency tool to copy the first 10 partials to the SPEECH layer
Invert the layer
Now you should be able to easily identify the wolf partials.
Select them individually with the frequency tool, adding each wolf tone to the selection.
Copy these to the NOISE layer, then phase invert the layer
Mute the SPEECH layer and the original speech will be heard and your fry should be gone.

When working with polarity in v9, you’ll have to make sure that when you copy the audio from layer to layer that BOTH layers are polarity matching. After that, you can then invert the layers and they will work as expected. I hope this gets fixed soon.

In v8, polarity works as expected You can leave the NOISE layer inverted and any audio pasted will become phase inverted automatically because the layer is inverted.