We propose a parallel sub-word recognition system (PSWR) as an alternative to the parallel phone recognition (PPR) system conventionally reported for language identification (LID) task. The sub-word recognizer (SWR) used in the PSWR system can be obtained from training data without phonetic transcription in any of the languages in the task. It is based on automatic segmentation followed by segment clustering and segment HMM modeling. The SWR can replace the front-end phone recognizer (PR) in the PPR system as well as in the PRLM and P-PRLM systems which constitute two other well accepted frameworks in LID system design. This allows easy expansion of these systems to a large number of languages without requiring tedious manually labeled training speech data in any of the languages in the task. On a 6 language LID task, using the OGI-TS database, we show that the PSWR system performs comparably to the PPR system, thus providing an efficient automatic alternative.