Study on the Reproduction of Vocal Fluctuations Using Box-Muller Method for Artificial Synthesis

Abstract

Human speech waveform has been synthesized in many ways. Today, text-to-speech synthesis has become the mainstreams in speech synthesis. However, this technology requires a large amount of data to make high-quality speech synthesis. For this reason, it is impossible to create a synthesized voice of many people. In this study, we aim to create a synthesized speech by focusing on the fluctuations of the vocal cords. By using Box-Muller method, we try to reproduce the fluctuations of the vocal cords. In the Box-Muller method, the average and variance values of the Gaussian function is used. In this paper, we extract the fundamental frequency and frequency variance value from the human voice. We expect that this result can help to reproduce the naturalness of the human voice.