A comparative study of explicit and implicit modelling of subsegmental speaker-specific excitation source information

摘要: In this paper, the explicit and implicit modelling of subsegmental excitation information are experimentally compared. For modelling, static dynamic values standard Liljencrants–Fant (LF) parameters that model glottal flow derivative (GFD) used. A simplified approximation method is proposed to compute these LF by locating closing opening instants. The approach significantly reduces computation needed implement model. linear prediction (LP) residual samples considered in blocks 5 ms with shift 2.5 Different speaker recognition studies performed using NIST-99 NIST-03 databases. case identification, provides better performance compared modelling. Alternatively, seem be providing verification. This indicates have relatively less intra inter-speaker variability. on other hand, has more What desirable Therefore, for verification task may used identification Further, both tasks complimentary state-of-the-art vocal tract features. contribution features robust against noise. We suggest can recognition.