RESUME

​Expressive Speech Synthesis : Research and System Design with Hidden Markov Models

​

Speech synthesis forms part of our everyday lives and has a wide number of practical applications (GPS, public-space announcements, electronic cards, etc.). However, the discrepancies that still remain between a synthetic voice and a natural human voice preclude its large-scale marketing. Its lack of naturalness is notably triggered by the inability of existing systems to deal with emotions, which are either not modelled at all or only in a very limited way.

Our project aims to develop an expressive speech synthesizer based on Hidden Markov Models (HMM). The advantage of HMM synthesis is that it allows an easy modification of prosodic and spectral parameters, without compromising the quality of the signal. Expressive HMM synthesizers that have recently been developed usually focus on Japanese which is characterized by restricted prosodic variations (fundamental frequency, duration, intensity). Moreover, they often make no use of explicit linguistic information and neglect the phonetic aspects. Our study will be divided into three objectives, aiming at improving the expressive synthesis quality: the improvement of automatic phonetic alignment of expressive speech corpora, their prosodic annotation and their phonetic processing with a view to  speech synthesis. The novelty of this project is that it sets out to use explicit linguistic parameters to improve the modelling of both prosody and phonetics.

• Music (I play the piano and the guitar)
• Reading
• Travelling
• ...

For more information, see my complete CV (updated in April 2013).