Keeping Voice, Removing Background Music How?

[Teac23]

HI

I have a long video clip and i would like to eliminate the background music so i am left with only the person's voice.

What software can do this?

Thanks for your help


Audio to Video Mixer

[guns1inger]

Posting a sample would be a good start.

Usually the request is for the opposite, and even then it is difficult. If this is a location recording then you are probably better off looking at ADR instead, however without a sample everything is guess work.

I have had some success using a variation on the phase inversion method to get the vocals only from songs, but it suffers form the same limitations as removing the vocals. That is, it relies on there being good stereo separation, and the vocals must be cleanly recorded straight down the centre, with everything else off centre.

[netmask56]

HI

I have a long video clip and i would like to eliminate the background music so i am left with only the person's voice.

What software can do this?

Thanks for your help


Audio to Video Mixer

If the recording is stereo and the voice typically placed in the centre then it is impossible to end up with a pure voice track. If you look at the algebra you will see why.
Any material panned or place to appear to come from the centre position is mixed with equal amounts to the left and right. If you sum the L and R you get L+R+C material. If you subtract L an R you get L+R with just a smidgin of the centre material depending on how accurate the original mix was and other inbalances.. Unfortunately there is no clean way of getting just the C information - it will always have L and R information mixed in. If you mix the left and right to produce a mono mix the vocal might be a couple of db higher and you can then use bandpass filtering techniques to more isolate the voice frequencies. In practise many recording studios employ techniques to prevent or minimise vocal removal by adding phase shift shuffle circuits to the vocal or common material. These are typically 45 degees plus and minus and some are dynamic in their operation. You get away with vocal cut if the replacement voice is more gutsy than the original so it masks the remaining original voice. I used to use an old QS encoder on my vocal tracks to screw up would be backing track rippers...

[guns1inger]

What you can do is get a voiceless track through phase inversion (or close to, given all the constraints) then use this track to remove everything but the voice. It is a manual process and takes time, but it can be done - again within the constraints of the method - i.e. there is no way to get just original voice track back, in the same way there is not way to completely remove just a voice track.

[netmask56]

The algebra of a stereo recording as follows (simplified)


Lt = L + ½C
Rt = R +½C

Sum them Lt + Rt = (L + ½C) + (R +½C) = (L+C+R) (mono release if required)

Subtract them Lt - Rt = (L + ½C) - (R +½C) = (L-R) (aurally identical to (R-L)

Where Lt = Left channel total
Where Rt = Right channel total

There is no way one can derive C

For example suppose we add (L-R) and (L+C+R) we have C+2L

etc

of course it gets even more difficult if C is phase shifted in the original mix by <180° typically ± 45°

Adobe Audition has a menu item to try and attempt centre channel extraction using a combination of bandpass and phase shift circuits - it sound dreadful! swishing phasing effect and noise gating etc.