Can You Fix a Damaged Mic in Post?

Frank Abraham produces Cooking Hawaiian Style, and sent an episode to work on. The good news is that there was a lot to love and, with glaring exception, it cleaned-up very nicely. The bad news is that one of the microphone inputs was badly distorted.

I'll talk about the minor issues first. Here's a few seconds from the original recording.

Image 1 - Voices are too low. Noise is too high.

Image 1 - Voices are too low. Noise is too high.

The voices are too low to start with (Total RMS = -35.45 dB) and the background noise (the two selected areas) is too high (Minimum RMS = -51.11 dB). If we simply crank up the volume, we will amplify the noise as well.

Happily, there was enough of a difference between the noise floor and the desired dialogue. After training the De-noiser, and making two passes at processing, the Minimum RMS dropped to -72.04 and stayed well-below -60 dB after processing to raise the volume to broadcast loudness standards.

Image 2 - A bit of room echo could be removed.

Image 2 - A bit of room echo could be removed.

There's a bit of room echo but processing it doesn't make a very dramatic improvement. Most people won't notice the difference, but the white spots and yellow dips show where echo is being reduced. (iZotope had not introduced ‘Dialog DeEcho’ yet.)

DeReverb
Now, that bad microphone was more of a problem. I've picked a section that has a bit of normal input, followed by distortion when another speaker comes in. Listen:

The audio on the right sounds like clipping, but its not. We’ve got a damaged microphone.

The audio on the right sounds like clipping, but its not. We’ve got a damaged microphone.

Image 3
You can see that I've selected an area to work on. The left side is just fine. On the right, the signal is squared off and sounds loud and rough; that's just bad. I'll leave the last 1.5 seconds unmodified (except for volume) for comparison.

My Waveform Statistics box isn't reporting any clipped samples, but I run the selected area through the De-clip module anyway, just to see what it does. It brings out some structure and fresh volume excursions, but doesn't manage its usual magic.

Conclusion: We're not seeing over-saturation of the pre-amplifier; the microphone itself is damaged. Like me, it has probably been dropped on it's head too many times. Note to self: Always test your set-up and monitor the sound while you're recording to catch stuff like this.

Image 4

Image 4

I run De-noise to clean up the background and Leveler to pull up the quiet left side while moderating the volume increase in the selected area. The distorted area has useless low frequency artifacts, which I remove below 50 Hz with a brick-wall high-pass filter. This picture was taken before applying the final Leveler finishing touch. Okay, time to compare to the original.

Wow. If you could get away with only playing this during the background noise in the middle of a trans-Pacific flight, most people might not notice. But, it's still not ready for prime-time. The episode is unusable. Damn.

David Satterlee