Indian Accent Detection using Dynamic Time Warping

摘要

This investigation aims at designing an accent detection system by recognizing various accents as an input. It aims to identify the accent of a person from his or her voice. The different accents that are detected by the system are the native Indian accents like Bengali, Gujarati, Malayalam and Marathi. The features used for detecting the accents are Voice Onset Time (VOT) and Mel Frequency Cepstral Coefficients (MFCC). These features were extracted for the spoken words having unvoiced stops p, , and k. Teager Energy Operator (TEO), which is a non linear energy tracking signal operator, is used for detecting the VOT. Thirteen MFCC features are extracted in the conventional way. The classifier used to detect the accented speech is Dynamic Time Warping (DTW) algorithm.