An Efficient Framework of Human Voice Verification for Robotic Applications

Abstract

Waves are considered to be used to decode the speech signal more efficiently. This study is an accessible and robust approach for obtaining voice recognition features. Here, we suggested a new text-related method for the identification of human voices (TDHVR) system, which utilizes the discrete wavelet transform (DWT) for low level feature extraction, Relative Spectral Algorithm (RSA) for denoising the voice signal and finally Additive Prognostication (AP) for estimating the formants. First, the proposed methods are used for voice signals, and then we construct a vector train function that includes the derived low level function and estimated formant parameters. The same technique is then applied for calculating speech signals and constructing a test feature vector. The Euclidean distance between the vectors will now be used to balance all vectors in order to distinguish the voice and voice. The simulated human voice would equal the educated person’s speech if the difference between two vectors is almost null. Computation results were compared with the LPC Scheme and revealed, that by using fifty preconfigured six voice signals, verification trials were carried out, and a best accuracy of approximately 90 percent was reached, the suggested methodology surpassed the current methodology.