Fast Estimation of Gaussian Mixture Model Parameters on GPU Using CUDA

摘要: Gaussian Mixture Models (GMMs) are widely used among scientists e.g. in statistics toolkits and data mining procedures. In order to estimate parameters of a GMM the Maximum Likelihood (ML) training is often utilized, more precisely Expectation-Maximization (EM) algorithm. Nowadays, lot tasks works with huge datasets, what makes estimation process time consuming (mainly for complex mixture models containing hundreds components). The paper presents an efficient robust implementation EM algorithm on GPU using NVIDIA's Compute Unified Device Architecture (CUDA). Also augmentation standard CPU version proposed utilizing SSE instructions. Time consumptions presented methods tested large dataset real speech from NIST Speaker Recognition Evaluation (SRE) 2008. Estimation proves be than 400 times faster 130 version, thus speed up was achieved without any approximations made formulas. Proposed also compared other implementations developed by departments over world proved fastest (at least 5 best published recently).