The members of the jury are: Xavier Binefa (UPF), Ana Barjau (UniversitatPolitecnica de Catalunya), Anders Askenfelt (Stockholm Royal Institute ofTechnology), Vesa Valimaki (Helsinki University of Technology), Rafael Ramirez (UPF).

Thesis abstract:
In this work we investigate new sound synthesis techniques forimitating musical instruments using the violin as a case study. It is amultidisciplinary research, covering several fields such as spectralmodeling, machine learning, analysis of musical gestures or musicalacoustics. It addresses sound production with a very empiricalapproach, based on the analysis of performance gestures as well as onthe measurement of acoustical properties of the violin. Based on thecharacteristics of the main vibrating elements of the violin, we dividethe study into two parts, namely bowed string and violin body soundradiation. With regard to the bowed string, we are interested inmodeling the influence of bowing controls on the spectrum of stringvibration. To accomplish this task we have developed a sensing systemfor accurate measurement of the bowing parameters. Analysis of realperformances allows a better understanding of the bowing control space,its use by performers and its effect on the timbre of the soundproduced. Besides, machine learning techniques are used to design agenerative timbre model that is able to predict spectral envelopescorresponding to a sequence of bowing controls. These envelopes canthen be filled with harmonic and noisy sound components to produce asynthetic string-vibration signal. In relation to the violin body, anew method for measuring acoustical violin-body impulse responses hasbeen conceived, based on bowed glissandi and a deconvolution algorithmof non-impulsive signals. Excitation is measured as string vibrationand responses are recorded with multiple microphones placed atdifferent angles around the violin, providing complete radiationpatterns at all frequencies. Both the results of the bowed string andthe violin body studies have been incorporated into a violinsynthesizer prototype based on sample concatenation. Predictedenvelopes of the timbre model are applied to the samples as atime-varying filter, which entails smoother concatenations and phrasesthat follow the nuances of the controlling gestures. These transformedsamples are finally convolved with a body impulse response to recreatea realistic violin sound. The different impulse responses used canenhance the listening experience by simulating different violins, oreffects such as stereo or violinist motion. Additionally, anexpressivity model has been integrated into the synthesizer, addingexpressive features such as timing deviations, dynamics or ornaments,thus augmenting the naturalness of the synthetic performances.