This paper gravitates on the spectrum channel allocation problem where each compounding node of a cognitive radio network is assigned a frequency channel for transmission over a given outgoing link, based on optimizing an overall network performance metric dependant on the level of interference among nearby nodes. In this context, genetically inspired algorithms have been extensively used so far for solving this optimization problem in a computationally efficient manner. This work extends previous preliminary research carried out by the authors on the application of the heuristic Harmony Search (HS) algorithm to this scenario by presenting further results and derivations on both HS-based centralized and distributed spectrum allocation techniques. Among such advances, a novel adaptive island-like distributed allocation procedure is presented, which dramatically decreases the transmission rate required for exchanging control traffic among nodes at a quantifiable yet negligible performance penalty. Extensive simulation results executed over networks of increasing size verify, on one hand, that our proposed technique achieves near-optimum spectral channel assignments at a low computational complexity. On the other hand, the obtained results assess that HS vastly outperforms genetically inspired allocation algorithms for the set of simulated scenarios. Finally, the proposed adaptive distributed allocation approach is shown to attain a control traffic bandwidth saving of more than 90% with respect to the naive implementation of a HS-based island allocation procedure.