*Or I don’t know how to call crosscorr() in MATLAB.
Anyway, a fine mini-project to dig into [ Code Composer Studio ] and Widrow-Hoff’s LMS is to isolate Chester Bennington’s voice, given that an instrumental version of New Divide is officially available on Amazon… [ New Divide | Instrumental ]
The laziest way of going about this is to subtract the two files. Now, they are not the same size!
So, if I just truncated the instrumental version, without considering any realignment, here’s what we get:
But a cross-correlation between just the left channels of each track, vocals and without, tells us that there’s an offset!
So, now that we know that one of the versions lags by 11098 samples (~0.25 seconds @44.1 kHz), we realign them, and here’s the cross-correlation (using a segment from 0:40 to 1:40)
There we go. Now, adaptive filters are the cooler of way of doing this, but I am not picking my mu and weights right. So, I need some reading and questioning to do!