Nicolas Obin, from IRCAM in Paris, will give a talk on "Analysis and Modeling of French Prosody for Artistic Purpose" on Monday March 16th at 13:00h in room 55.410 (4th floor of Tànger building).
Abstract: Over the past decade, voice and speech technologies have received a significant increase of attention in multi-media and artistic applications. In particular recent improvements in speech synthesis make it possible to generate natural speech from text. However if the generated speech sounds "acoustically" natural (coarticulation, phonem level) it could appear unnatural according to the expected "way of speaking" and is most of the time repetitive and boring. This is due to the fact that speech synthesizers did not use to take into account such "way of speaking", i.e. the prosody of speech. Thus prosody has become a main focus in speech research over the past few years. At the acoustic level prosody is related to the macro variations of a set of acoustic parameters of speech (fundamental frequency, local speech rate, intensity, voice quality,...) over time. Such variations are used by a speaker to convey meaning and emotional states. In this talk I am going to present some aspects of my research work on prosody applied to French langage. This work aims to model prosody with high-level linguistic features (syntactical, semantic, discursive) in different speaking styles. I will first introduce some concepts about prosody and linguistic; thus I will present the global framework of prosody modeling. Then I will present some of my work on the analysis of French prosody: syllable-based prominence detection, speaking style analysis and finally introduce my proposed prosodic model based on parametric multi-level context-dependent analysis. I will also present some aspects of my concerns in the artistic usage of this type of research.