La Salle R&D has finished the Project CreaVeu. The main outcome of this project is the development of a new tool for voice presets creation in a Text-to-Speech (TTS) synthesis framework. Our previous experience in Unit-Selection-based TTS (US-TTS) systems has led us to define a new strategic research line in order to achieve more flexible TTS systems which provide diversity of voice characters using a unique voice database. This tool enables personalization and fine tuning in terms of specific low-level voice parameters (i.e. speech rhythm, mean F0, vocal tract, voiced/unvoiced balance, etc.) through the use of Harmonics plus Noise Model (HNM). The user can easily interact using a simple GUI to load and modify existing voice presets or to save new voice preset banks.
Creaveu tool integrates our previous results carried out in US-TTS systems (i.e. optimization of the unit-selection process, expressive prosody modeling, etc.) with new methods of signal transformation by using web service-based technology and HNM. Our aim is to continue researching in the field of signal processing techniques with the aim of improving versatility and flexibility of US-TTS systems. In this sense, we are working on new transformation parameters (i.e. vocal effort) that can have a deep impact in terms of expressivity of voice characters.