This shows you the differences between two versions of the page.

Link to this comparison view

articulatory_singing_voice_synthesis_for_music_production [2014/04/04 13:37] (current)
bziolko created
Line 1: Line 1:
 +Martin Swanholm, Thomas Barslev (Sweden)
 +**Articulatory singing voice synthesis for music production**
 +Articulatory speech synthesis is a type of physical modeling synthesis tailored to model human speech production. This method is known to produce more natural-sounding and usually more intellegible results than traditional speech synthesis methods such as concatenative synthesis, at the expense of being more computationally intensive. With advances in computational power, real-time synthesis using this technique is becoming increasingly relevant. In this project, a model for articulatory speech synthesis is applied to the domain of musical singing voice synthesis. Benefits of using this model include flexibility and expressiveness. We investigate what model parameters are important in a musical context and how the speech model can be extended to incorporate additional parameters that address emotion, special vocal effects and other forms of musical expression. Parameter value ranges are critical to expressiveness, naturalness and usability. A number of methods for determining useful ranges are investigated, including user tests and subjective listening tests performed with composers and music producers, as well as analysis of a capella song recordings of professional singers. In the analysis, relevant time-varying features are extracted using digital signal processing techniques, and model parameters can be mapped accordingly. Preliminary results are promising - in particular in terms of expressiveness, allowing a wide spectrum of vocal qualities ranging from very artificial synthetic instruments to human-like singing voices. Audio examples are presented to demonstrate results.
Copyright © XXII PVC Organizing Committee 2013. All Rights Reserved.