Benchmark for Speaker Identification Using Glottal Source Parameters in Hindi Speakers

Abstract

Studies concerned with establishing parameters for speaker verification are important because of the legal ramifications and because of the forensic involvements associated with the application of these studies. Success of identifying the speaker depends on extracting speaker-dependent features from speech signals that can effectively distinguish one speaker from another. It is not known as to what percent matching would indicate similarity/ dissimilarity of speaker, or benchmarking of various features is not established In this context the aim of the present study was to determine the benchmark for speaker identification using glottal source parameters in direct recording condition. Ten normal Hindi speaking male subjects in the age range of 21 -38 years participated in the study. The material used was nine commonly occurring, meaningful Hindi words containing the long vowels la:/, Ii:/, and /u:/ in the word-medial position embedded in sentences. The vowels were displayed as waveform and were acoustically zoomed to extract the source and filter parameters using Acophone I and SSL software (Voice and Speech Systems, Bangalore). Glottal source parameters open quotient (OQ), leakage quotient (LQ) and speed quotient (SQ) were extracted in I 0 steady state point of each of the vowels. The results of the present study showed that the glottal source doesn't remain the same even in normal mode of speaking. Hence these parameters don 't serve as a good measure for speaker identification. In general it could be concluded that OQ*LQ, OQ*SQ, and LQ*SQ cannot be considered as an efficient parameter for speaker identification in field conditions in Hindi speakers.