Abstract—This paper presents a novel approach to language identification (LID) which is unique n-phone ranking (UnPR) calculated from p-best phone recogniser outputs. The idea underlying this approach is to create a list consisting of a set of unique n-phones considered to be the approximate identity of a certain language. Thus, every language in the target set has its own list. These lists support the process of distinguishing one language from another. The novel part of the proposed approach lies in the procedures for generating such lists and utilising them to perform LID tasks. Compared to our previous work, the added value of the work presented in this paper is taking phone lattices into account to investigate whether the updated UnPR can improve the LID performance. Besides the UnPR from our previous work, parallel phone recognition followed by language modelling (PPRLM) is also used as a baseline system. Furthermore, a fusion combining the UnPR system with a number of configurations is carried out. The experiments show that the proposed approach complemented with lattices improves LID accuracies. The amount of time spent is also significantly faster, which is ~9.5 times faster than the PPRLM system.