Subtracting a similar but not identical audio?

Hey there. Basically what I’m trying to do is, some sort of… combination between subtracting and noise removal, in an automated way?

I was recording commentary of a game while the game was playing and did not realize how much the game audio also picked up on my microphone. I have the game audio separately ripped from the video, so is there any way to use a noise removal-type function to compare the two tracks and subtract the game audio from it that way? Naturally, the invert and subtract method wouldn’t work in this.

To add a wrinkle to it… because of some issues with the Audacity commentary not recording at the right speed, I had to re-sync it by manually stretching it, so it may be off by a tenth of a second in places.

Any thoughts on this or am I out of luck? I’m just trying to salvage what I have, I really should have done better test recordings before I got into this, I know.

Thanks for any help.

I have some cold water left.

Noise Reduction and tools like that all depend on an interference that does not change. Quiet fan noise from a computer counts, but a car starting up outside and driving away does not. Music certainly doesn’t count. The musical notes are constantly changing.

You are correct that the simple cancellation process isn’t going to work. That depends on the two tracks being surgically identical to each other except for the additional voice in one. In practical terms, people insist on using MP3 and other compressed sound formats which shoots you right in the head because MP3 distortion and compression problems guarantees a bad match – or cancellation.

You run into a lot of the same problems of people doing simple voice recording or singing. Big fuzzy headphones are required to keep the parts of the performance separate so you can do post production on them later.

I have a Logitech A00008 USB headset that seems to work OK. It has good isolation and good noise rejection in the microphone, although that does make me sound a little “tight” with no bass.

I thought I had a sample here, somewhere.

This is me on a broadcast microphone as a contrast.


Could you perhaps submit a sample?
Are the two tracks both mono?

The two channels don’t have to be 100% identical (the game sound I mean) to improve it with the vocal remover.
It depends on the frequency. A 100 Hz tone will be silenced when both channels are in sync.
Those are the reduction values for the various shifts:
0 °
-1.#INF (silence)
7.5 °
-17.6674 dB
15 °
-11.6655 dB
22.5 °
-8.17473 dB
30 °
-5.7195 dB
37.5 °
-3.83742 dB
45 °
-2.32262 dB
52.5 °
-1.06531 dB
60 °
-0 dB

(Same values for -60…-7.5 °)

This means that the common sound will still be attenuated when the phase difference is less than 60 °.
However, for our 100 Hz tone, this would be mere +/- 2 ms.
I would probably try the Noise removal first.
Take the profile from the game sound and apply it to the voice track.
The most important value is perhaps the sensitivity level.
Since the game sound is much quieter in the commentary track, it should be accordingly low (negative).