Speaker Identification in Vehicular Environment using GMM and BiLSTM

Speaker recognition in the vehicular environment is becoming increasingly significant with the advancement of smart and connected devices. Noisy conditions in the vehicular environment pose a pressing challenge to speaker identification tasks. Thus, a robust speaker recognition system is proposed that is not susceptible to a drop in performance when subjected to various SNR levels. This paper analyzes the performance of speaker identification using GMM and BiLSTM in noisy environments. The model is trained to identify speakers from the TIMIT speech corpus. The paper also explores data augmentation techniques, namely, time stretching and time rolling. On training and testing the proposed speaker identification system with TIMIT and custom dataset, results show that more data for both training and testing purposes can eventually improve the performance of the system.