Musical and Phonetic Controls in a Singing Voice Synthesizer

have directed the project in Hamamatsu (Japan) and have visited us regularly in Barcelona. The rest of the people in the Music Technology Group have also to be mentioned, they have been always available to give us a hand when needed. In addition, I want to express my gratitude to Luis Vergara who has directed this thesis from the Polytechnics University of Valencia. Barcelona) in collaboration with Yamaha Corporation. The Music Technology Group, MTG, is a research group working on signal processing techniques for musical production and for other multimedia applications. Apart from pursuing the development of spectral audio models, MTG is dedicated to sound models for synthesis, the processing of audio based content and other issues related to Music Technology. On the other hand, Yamaha Corporation manufactures all kinds of musical instruments and professional audio equipment for professionals and amateur enthusiasts. From its base in Hamamatsu City, southwest of Tokyo, the company is also a leading producer of audiovisual products, semiconductors and other computer-related products, electronic equipment and specialty metals. The aim of the Daisy project is to synthesize a singing voice from a musical score. That is to say, from a given musical melody and a given lyrics in a particular language (English and Japanese in our case) our goal is to obtain an output sound as though a real singer was performing a song. Of course, this is not an easy thing, to say the least. However, with the accumulated knowledge in different fields, the use of new technologies and the increasing power of computers, this objective has become achievable nowadays. The artistic and technical disciplines relevant to this project cover an impressive variety of fields: sound recording and reproduction, music performance, music perception, phonetics, computer programming, digital signal processing… We can say that this is a really multidisciplinary enterprise. This research project presented here is a continuation of an automatic singing voice impersonator application for karaoke developed by the Music Technology Group [Cano, Loscos, Bonada, de Boer, Serra, 2000]. That system morphed in real time the voice attributes of a user (such as pitch, timbre, vibrato and articulations) with the ones from a prerecorded singer. 2 Because of my education as a musician and as an engineer, I have always been willing to work in an area in which I could apply my knowledge in both fields. And from the first time I heard about the …