The sound inventories of the world's languages show a considerable extent of symmetry. It has been postulated that this symmetry is a reflection of the human physiological, cognitive and societal factors. Although the organization of the vowel systems has been satisfactorily explained for smaller inventories, the structure of the consonant inventories is an open problem since 1939. We reformulate the problem in the light of statistical physics, more precisely complex networks, and observe that the distribution of the occurrence and co-occurrence of the phonemes (consonants and vowels) over languages are scale-free. The co-occurrence network exhibits strong community structures, where the driving forces behind the community formation are the human articulatory and perceptual factors. In order to validate the above principle, we introduce an information theoretic definition of these factors - feature entropy and feature distance - and show that the natural language inventories are significantly different in these terms from the randomly generated ones. A preferential attachment based growth model can lead to the emergence of similar topologies as that of the real networks. Furthermore, in a separate study, we observe that spectral analysis of the co-occurrence network of consonants helps us in the induction of linguistic typologies.