Voice Recognition Technology Slowly Advancing

The problem with using voice recognition as a biometric is its slow processing speed.

However, researchers at North Carolina State University are attempting to
improve the speed of speech authentication, without sacrificing accuracy.

According to the research paper, “Joint Frame and Gaussian Selection for Text Independent Speaker Verification,” the research is based on Gaussian selection, a technique applied in the GMM-UBM framework to accelerate score calculation. The researchers modified the method known as sorted GMM (SGMM). SGMM uses scalar-indexing of the universal background model mean vectors to achieve fast search of the top-scoring Gaussians. They extended this method by using 2-D indexing, which leads to simultaneous frame and Gaussian selection. The results on the NIST 2002 speaker recognition evaluation corpus indicate that both the 1- and 2-D SGMMs outperform frame decimation and temporal tracking of top-scoring Gaussians by a wide margin.

Voice recognition technology can be used to prevent ID theft and data protection for the government, financial, healthcare and telecommunications sectors.