Singing voice transformation system

• Main Authors: Bullen, John Michael A., Carandang, Wilfredo M., Jr., Villamin, James Vincent A.
• Format: text
• Language: English
• Published: Animo Repository 2006
• Subjects: Speech processing systems; Digital signal processing; Signal procesing > Digital techniques; Electronic music; Computer Sciences
• Online Access: https://animorepository.dlsu.edu.ph/etd_bachelors/8958
• Institution: De La Salle University

Speech processing is a broad field of study with a variety of applications such as speech recognition, analysis and synthesis. One of the less explored fields of study is its application on the singing voice. Singing voice transformation deals with modification of the characteristics seen in the frequency spectrum of an input singing voice signal such as the pitch, amplitude and spectral shape so that a transformed voice signal is given as output. The thesis is direct application of singing voice transformation using various digital signal processing techniques. The application is a singing voice transformation system where the characteristics of the user's singing voice such as the fundamental frequency, amplitude, and spectral shaped are modified such that they resemble the characteristics of the target voice while preserving the quality of the transformed voice in terms of continuity, smoothness and its overall perceptibility. Quantitative and qualitative tests are done to asses the performance of the system in transforming the fundamental frequency, amplitude and spectral shape of the user's singing voice in terms of quality and accuracy of the transformed voice. Specifically, the Mean Opinion Score (MOS) method of subjective testing proves that the system's performance on the male-to-male, male-to-female, female-to-female, and female-to-male voice transformation is satisfactory. From a scale of 1 to 5, with 5 being the highest score, the system has a performance rating of 3.7 in terms of quality and 4 in terms of accuracy. Moreover, the most notable performance of the system is on complete transformation of the user's voice into the target voice, i.e. the fundamental frequency, amplitude, and spectral shape characteristics of the user's voice are all transformed to resemble those of the target voice's characteristics. The system is able to perfectly transform the user's voice into the target voice. Furthermore, the quantitative tests prove that the system is able to perform fundamental frequency