Development of Language Identification using Line Spectral Frequencies and Learning Vector Quantization Networks

Abstract

Language identification system has become a very active research nowadays due to the need of intercultural human communication. This paper proposed a Language Identification System using Line Spectral Frequencies (LSF) and Linear Vector Quantization (LVQ) network. LSF was used due to its robustness compared to normal linear predictor coefficients (LPC), while LVQ was used due to its low complexity. Three languages, i.e. Arabic, Malay, and Thai, for both native male and female speakers were recorded at IIUM Recording Studio. Several experiments have been conducted to find the optimum parameters, i.e. sampling frequency (8000 Hz), LPC order (18), number of hidden layers (300), and learning rate (0.01). Results show that our proposed system is able to recognize the trained languages with the recognition rate of 73.8%. Further research could be conducted to improve the performance using different features, classifiers, or using deep learning neural network.