Performance Evaluation of Text Independent Automatic Speaker Recognition using VQ and GMM
Sumita Nainan, Vaishali Kulkarni · 2016
With rapid strides in web based applications, the need of the hour is to have faster and less complex methods to authenticate access for critical applications. The challenge is in curbing the rampant rise in fraudulent means of Acquiring Identity. Spoofing, flexibility and Accuracy are the areas to be addressed for Cyber Security. This paper attempts to address this issue where Text Independent Automatic Speaker Identification using Vector Quantization (VQ) and Gaussian Mixture Model (GMM) has been done. The VidTimit data base has been used for the same. An accuracy of 92% is achieved using GMM as compared to 72% for VQ method. We however propose that as no single modality offers 100% accuracy, for Text Independent Speaker Recognition System, Multimodal System needs to be considered for achieving accurate authentication and verification of individuals.