Hi Achilleas,
Thank you for this. Incidentally, this code matches the peak detection algorithm used in gr-dab too (and IIRC, gr-dtv dvbt Receiver too).
Though this code doesn't explicitly remedy the problem I was facing, it still has a fairly good implementation (post data-smoothening). I'll use the understanding from this implementation into my own implementation too.
As for the smoothening problem, Marcus's prompt about normalizing the correlations with signal energy worked for me. I'll keep you posted. :)