| Abstract [eng] |
Voice biometric authentication method based on the fact that the human voice itself has a significant amount of information: the language (English, Russian, Lithuanian), speech pathology, sex, emotional status, age. Therefore, as the fingerprints of human voices are different. Although the speaker identification accuracy is lower for fingerprints or retinal scanning, but the speaker recognition hardware is much cheaper, just a simple microphone is needed. This work will review the experiments and studies done in this field. As well as an overview of the main principles and methods of speakers identification. Their operational nature, uses, and the accuracy recognition. So the experiment was developed, which aims to speaker identification. Program operation consists of three main steps: data collection, training and testing. The analysis of the theory resulted in creating a plan of the experiment, which has been realized. Created an authentication program, with the operating principle: taken audio files are processed, the necessary experimental word "one" is cut , and a new sound file is handled (noise and silence reduction, and DC offset elimination). Then, out of testing phase, during which the language model is trained, a VQ code book for each standard. Later, the code book will be used for speaker recognition. Testing at the same loaded the sound files in succession, and the use of VQ code book is calculated using Euclidean distance, the standard test with which the speaker has the smallest distortion is recognized as a speaker. |
