Improvement of speaker verification for Thai language

There are many strategies proposed for speaker verification (SV) system, both in text-dependent (fixed-text) and textindependent (free-text) domains. To convey an appropriate algorithm for Thai speech, several consecutively improvement methods are compared in this paper including the dynamic time warping (DTW) matching and Gaussian mixture model (GMM) based systems. We firstly developed a system based on the conventional scoring algorithm. This system is improved by the incorporation of many scoring algorithms such as the cohort normalization, the global speaker model (GSM), and a new approach, namely, global anti-speaker model (GASM). Experiments are set up for Thai numeral speech and the results show an improving tendency of each algorithm.