Language-Identification-using-Time-Frequency-Image-Textural-Descriptors-and-GWO-based-Feature-Selection

An ability to categorize and recognize a spoken language is an essential task in a multi-lingual society like India. Language identification (LID) is the process of identifying the language that is being spoken by some unknown speaker using a given speech sample. In this project, we have designed a hardware implementation of language identification using NVIDIA Tesla K80 (GPU) in real time. The frequency based features i.e., prosodic features are different for every language and also follows a certain patterns, which helps to recognize the language of an unknown speech. The proposed LID approach consists of four main stages. In the first stage, an audio sample is converted into a spectrogram visual representation which is a representation of the band of frequencies of a signal with respect to time. In the second stage, Rotational Invariant Complete Linear Binary Pattern (RICLBP) is used to extract the features from the spectrogram image. In the third stage, using Grey Wolf Optimizer (GWO), irrelevant and redundant features are removed and only optimal features are selected from the data set hence it helped to construct the classification model and the performance of the classifier is optimized. In the final stage, using Deep Neural Network, the pattern from the selected features are recognized by the trained classifier and classifies the unknown language sample to a known category.