Title: Text-independent Speaker Recognition of Mandarin by Big Data Technic of GMM and VQGMM
Abstract: Speaker recognition has been long a popular research topic as to identify a specific speaker by his free speech of text-independent. This study uses Mel-frequency Cepstrum technic to simulate and extract speakers features; uses big data technic of Gaussian Mixture Model, and Vector Quantization Gaussian Mixture Model to find out the impact factors that affecting speaker identification hit rate. Research result showed training identification hit rate saturated when Gaussian mixture numbers reached specific level, then start to decreasing. Research result showed VQMM owns triple performance than GMM at training data sets. Research result suggests upcoming scholars do not take GM number as the sore tool to achieve identification hit rat.