Building Speaker Identification Dataset for Noisy Conditions

Abstract

Speech signal processing plays a crucial role in any speech-related system whether Automatic Speech Recognition or Speaker Recognition or Speech Synthesis or something else. Burmese language can be considered as an under resourced language due to its linguistic resource availability. For building Burmese speaker identification system, the sufficient amount of speech data collection is a very challenging task in a short time. In order to get higher data size, this paper analyzes that the getting higher duration of speech data actually combining with various noises encountering in our surroundings. For increased noisy state speech data, we also used the voice activity detection (VAD) technique to acquire only the speaker specific information. For feature extraction, we used MFCC, Filter Banks and PLP techniques. The experiments were developed with i-vector methods on GMM-UBM together with PLDA and presented the performance of different data set in the form of EER with two models trained on clean and noisy data to prove that the developed speaker identification system is noise robust.