the analog way of doing it would be as follows:
1. Use one low-pass filter per channel to isolate all frequencies below vocal range.
2. Use one high-pass filter per channel to isolate all frequencies above vocal range.
3. Use one band-pass filter per channel to isolate all vocal range frequencies.
4. Feed the isolated vocal range frequencies into an A-B (Left minus Right) filter to remove all common information (vocals are common in both channels).
5. Sum the above filtered (A-B) information + Low-pass + High Pass information back together for each channel.
Sounds a little complicated, but its not if you are a DIY'er.
Downsides:
Will not work on mono recordings.
Stereo imaging will be lost in the band-pass frequencies
Any other common information (besides voice) in the band-pass frequencies will be lost.
There may be a slight bit of voice left (2-5%) if the recording is not perfect, of if resistor / capacitor tolerances are not good.
