Rychlá estimace parametrů model Gaussovských směsí za využití GPU a architektury CUDA

Gaussian Mixture Model (GMM) statistics are required for maximum likelihood training as well as for adaptation techniques. In order to train/adapt a reliable model a lot of data are needed, what makes the estimation process time consuming. The paper presents an efficient implementation of estimation of GMM statistics on GPU using NVIDIA's Compute Unified Device Architecture (CUDA). Also an augmentation of the standard CPU version is proposed utilizing SSE instructions. Time consumptions of presented methods are tested on a large dataset of real speech data from the NIST Speaker Recognition Evaluation 2008. Estimation on GPU proves to be 100 times faster than the standard CPU version and 30 times faster than the SSE version assuming more than 256 mixtures, thus a huge speed-up was achieved without any approximations made in the estimation formulas. Proposed implementation was also compared to other implementations developed by other departments over the world and proved to be the fastest