Hi guys,

I recently shot some outdoor interviews in the forest where cicada noise was quite bad.

Can anyone offer any recommendations on how to reduce the noise in post? I've tried tweaks with a simple 3-band EQ which worked to an extent but the noise is still distracting.

I have access to Audacity and Izotope RX3 if that helps.

Any tips appreciated!
Thanks

Please post a sample clip.

Hi Greg,

Here's some raw audio, no post processing has been done:

https://dl.dropboxusercontent.com/u/7711130/temp.wav

Thanks

Hi Jody,

Well, I should have asked you to post a clip that had more cicadas (without speech) to use as a noise sample. But I think I made out OK with this one.

First, I looked at the frequency content of the clip, looking at the area around 1.4 seconds (where there were some cicadas only) and it appeared that the cicadas were a hump starting around 4.5 kHz and going upward. Much too broad a frequency range to remove with any LPF or notch filter.

I listened to the file through a LPF, to confirm the lowest frequency occupied by the cicadas, and it was approximately 4.5 kHz (as I had suspected from the freq. content graph).

I didn't want to run a wide-band noise reduction on the file, because the result would have been unnaturally quiet. So I selected that short area around 1.4 seconds, and ran that through a HPF. 100% at 4.0 kHz and above, 0% at 3.0kHz and below. That left mostly the cicadas plus a bit of other HF noise. I then used this part of the file as a noise sample for the adaptive noise filter. (That way, I will adaptively remove noise above 3.0 kHz, so all the lower frequency ambience will remain, giving a less artificial sound to the result.)

After getting the noise sample, I remembered to undo the HPF around 1.4 seconds, and then applied the adaptive noise filter to the entire file.

If you listen carefully, you will hear the cicadas slightly gating on and off as the talent speaks. It's pretty subtle and sounds to me just like a bit of hiss. It's tough to avoid that, since the frequency range of the cicadas overlaps the range of the speech. Since the cicadas are gone during areas without speech, it's fairly unobtrusive and not especially recognizable (at least not to me).

If you try this technique, you may want to play around a bit with the various settings.

What do you think?

Wow, the difference is incredible. I expected that a slight improvement was possible, but not to that extent.

Thanks very much for the help. I'll have a play around myself and try to replicate those settings.

What audio editor do you use?

For this file I used Cool Edit Pro, which is long since defunct. I have access to Audition (which replaced CEP) at work, but haven't bothered to install it here at the home studio. I also have Diamond Cut Audio here, and it has some tools that CEP doesn't, but I find the CEP UI so much more intuitive that I use DC only for really difficult and special processing.

Do you have access to something comparable? The adaptive noise filter is the important part for this particular file.

In retrospect, I ran your file at 48 k as you sent it. Based on past experience, I believe I could have gotten somewhat better results if I had converted it to 96 k before processing (and then back again afterwards). Also, increasing the number of FFT points in the adaptive noise filter would probably help some (I think I had it set at 2,048). And I used just 12dB of NR; you could lower it further if you want; the tradeoff would be that you might hear the cicadas pumping more obviously than in the version I uploaded.

Greg, wow. Big kudos to you for pulling out that much cicada without badly distorting the voice! Sure, she's missing a little of the "brightness" of the original recording, but without hearing the comparison I doubt anyone would balk at that. Jody, if you get to picture-lock and still need help, you might consider hiring Greg on spec ; )

For this file I used Cool Edit Pro, which is long since defunct. I have access to Audition (which replaced CEP) at work, but haven't bothered to install it here at the home studio. I also have Diamond Cut Audio here, and it has some tools that CEP doesn't, but I find the CEP UI so much more intuitive that I use DC only for really difficult and special processing.

Do you have access to something comparable? The adaptive noise filter is the important part for this particular file.

In retrospect, I ran your file at 48 k as you sent it. Based on past experience, I believe I could have gotten somewhat better results if I had converted it to 96 k before processing (and then back again afterwards). Also, increasing the number of FFT points in the adaptive noise filter would probably help some (I think I had it set at 2,048). And I used just 12dB of NR; you could lower it further if you want; the tradeoff would be that you might hear the cicadas pumping more obviously than in the version I uploaded.

Thanks for that. I've had a play around with the file in Audacity and haven't been able to replicate your result but I'm getting slightly closer. I'll keep at it!

Thanks again.