Language Identification Using PLDA Based on I-Vector in Noisy Environment

This paper investigates a new language identification technique based on i-vector and PLDA in noisy environment. Previously various technique were employed to find identity of language. In high dimension and noisy environment, existing techniques became too complex and degraded their performance. We have developed a new technique based on i-vector and applied PLDA over internally centered i-vector for noisy data set to make it robust and efficient. For clean speech, the average equal error rate of traditional GMM system and proposed system are 19.002% and 5.4278%, respectively. The performance of language identification system is degraded significantly in noisy environment. The average equal error rate in noisy environment is 37.493% for baseline system and 30.7764% for proposed system. In this paper, the speech enhancement method (spectral subtraction) is used for noise suppression. After enhancing speech, the baseline and the proposed system shows output by 26.525% and 16.695%, respectively in term of EER. Experimental result shows the robustness of our proposed language identification technique over existing