MATLAB based speaker verification system ¨C generative modelling

• Main Author: Hong, William Guang Yu.
• Other Authors: Gan Woon Seng
• Format: Final Year Project
• Language: English
• Published: 2010
• Subjects: DRNTU::Engineering::Electrical and electronic engineering::Computer hardware, software and systems
• Online Access: http://hdl.handle.net/10356/40127
• Institution: Nanyang Technological University

The distinctive features of a person’s voice enable us to verify the caller identity. There are several voice recognition and voice verification system developed. One is the text-dependent which must match with a specific preset text by the programmer. The other is text-independent voice verification systems, the user or the speech input need not have a specific text to say or follow. In the paper and the author’s FYP is about creating a text-independent voice verification graphic user interface (GUI) for use. Voice verification starts by taking a voice input speech and extracting the important features, a process also known as parameterisation. Once an unknown voice input is fed in, the system will extract important features in a voice within a 20ms window with 10ms overlapping for each window. Most of the redundancies will be discarded and the important features will be stored as feature vectors or .mfcc files. After doing so, a Gaussian Mixture model- Universal Background Model (GMM-UBM) is either created beforehand, or during the speaker verification. Normally it will be created beforehand as if there are a lot of speakers used for this background model it will take a long time to create. After that, the feature vector file is enrolled and a .gmm file is trained out. From there, the system will compare scores between the unknown speech input and whatever default voice sample the user chooses. By following these steps, the author has created a simple speaker verification GUI to be hopefully put to use into companies where the extra security is needed.