Development of accurate automated language identification model using polymer pattern and tent maximum absolute pooling techniques

Abstract

Various language identification tools and methods have been used in the real world. These applications can detect language using text or images. However, there is no speech-based language automated identification tool available. Therefore, many studies have been presented to overcome this problem. This work presents an automated high accurate language identification model and developed a new corpus for language identification. The developed language identification model uses two novel methods: (i) polymer pattern (PP) and (ii) tent maximum absolute pooling (TMAP). These methods help to extract both low- and high-frequency features. In order to choose the most informative features, a threshold-based iterative feature selector is presented. The proposed PP- and TMAP-based model has attained an accuracy of 97.87% and 99.70% using our newly developed and VoxForge datasets, respectively, with kNN classifier with tenfold cross-validation.