Assamese Dialect Identification System Using Deep Learning

The study describes a deep learning-based method for identifying Assamese dialect from speech.Dialect identification systems offer a wide range of uses in a multilingual country like India.The study examined the speech samples of four significant regional dialects of the Assamese language.These include Central Assamese, spoken in and around the Nagaon district; Eastern Assamese dialect, spoken in Sibsagar and nearby districts; Kamrupia dialect, spoken in Kamrup, Nalbari, Barpeta, Kokrajhar, and some parts of Bongaigaon district; and Goalporia dialect, spoken in Goalpara, Dhubri, and parts of Bongaigaon district.To achieve this objective, a database was constructed, consisting of ten speakers (five males and five females) who represented each of the mentioned dialect regions.After obtaining the spectrograms, further processing is performed, including framing and windowing, to prepare them for input into a Convolution Neural Network (CNN).The framed and windowed spectrograms are then fed into the CNN for analysis and processing.This analysis makes use of a mix of spoken utterances from a binary combination of various dialects of Assamese Languages.The performance evaluation of the system is conducted using various metrics such as accuracy, precision, recall, F1 score, and the confusion matrix.These measures are employed to assess and measure the system's effectiveness and overall performance.The maximum accuracy achieved for the data set tested is 99.4% for Eastern vs Central Dialect and 99 % for Kamrupia vs Central Dialect.