Isolate sounds

[Kathy Smith]

Hi,

I'm pretty sure this is impossible but I'll give it a try. I recorded an interview but it's in an environment where other people are talking and even though the desirable voice is audible I would love to be able to isolate it from the background voices to some degree. Is this doable in any way?

[Bruce Watson]

Originally Posted by Kathy Smith

I recorded an interview but it's in an environment where other people are talking and even though the desirable voice is audible I would love to be able to isolate it from the background voices to some degree. Is this doable in any way?

Fix it in post? Probably not. The sounds you want to eliminate share frequencies with the sounds you want to keep. You can't eliminate one without eliminating the other.

Signal to noise ratio with audio is all about microphone placement. If you need greater signal to noise ratio, in general, get the mic closer to the signal.

For example, people do this kind of thing all the time with trade show coverage. An omni reporter's mic at 10cm from the mouth, and my recording makes the speaker's voice clearly audible over the din. Sports casters do this all the time at truly loud sporting events. You can't hardly watch a local newscast without seeing an in-the-street interview done this way.

[Rick Reineke]

In some instances, 'spectral repair' software (iZotope RX & SCS SpectraLayers for instance) can 'eliminate' unwanted noises, but I'm not aware of any 'auto' or simple procedure like din-type noise reduction, and would likely be time consuming... not to mention the time it takes to gain proficiency with the SR software, i.e., significant learning curve.
The results would be uncertain as well, depending on many variables.

[Greg Miller]

I agree with Mr. Watson. The wanted and unwanted frequencies overlap, so frequency-dependent filtering won't work. (If the wanted voice is a low male voice, and the unwanted voices are high female voices -- or vice versa -- you might have a small bit of luck, but I wouldn't bet money on the result, because there is still a significant amount of frequency overlap.)

If the unwanted voices are much lower in level, you might get a slight improvement with downward expansion. But if they're already significantly lower in level, you probably wouldn't be asking this question.

[Roger Gunkel]

There are only three things I can think of.

1 Use a volume curve on the audio track and manually increase the volume of all the bits you want or reduce the volume of the bits you don't.

2 If there is a noticeable difference between the interview and the background audio, it might be possible to use a noise gate, with the threshold set to open at the volume of the required audio and close on the inbetween audio. It would need to be set with a very fast attack and release time, but might do the job.

3 Manually chop up the audio to only keep the required interview audio, then add a low level sample loop of the background noise to make the gaps inbetween less obvious, or go to my first suggestion and use the volume curve to get the same effect.

As the others have said though, the best way to get audio correct, is to record it using the right equipment in the first place. Always easy to be wise afterwards though.

Roger

[Andrew Smith]

A sample of the audio might help.

Andrew

[Graham Bernard]

Originally Posted by Andrew Smith

A sample of the audio might help.

Andrew

Yes, I echo your request.

G

[Kathy Smith]

Hi,

I just want to update this threat. I e-mailed the sample to Andrew and he was able to help me out. Thank you everyone!

[Don Palomaki]

What were the tricks he applied?

One thought, if it was recorded with a stereo mic and the subject is centered on on axis, mixing the left and right channels down to mono might help bring up the on axis sound relative to the off axis sounds.

[Andrew Smith]

First trick was to get the original unmodified (clean-up attempted) wav file. :-)

The first part was a spectral noise-reduction repair in Izotope RX5 where I selected all the background noise samples (in-between the subject talking) as my target sample for what had to be removed. This got the bulk of it.

The second part was a further noise reduction using the 'voice' part of the noise reduction plugin. This refined the cleaning up of the audio just that little bit further.

I'm just glad that we were able to sort out the audio for her.

Andrew