Section: Contracts and Grants with Industry
The team is involved in the management of the regional CPER contract. In particular, Christophe Cerisara is co-responsible, with Claire Gardent, for the CPER MISN TALC.
Investigation of speech production (MODAP)
The acquisition of articulatory data represents a key challenge in the investigation of speech production. These data could be used either to improve the naturalness of talking heads, or to add further information in automatic speech recognition. We thus initiated cooperation with Equipe IMS from SUPELEC and EPI Magrit to capture and exploit articulatory data.
This project relies on an articulograph AG 500 (base on the tracking of electromagnetic sensors) developed by Carstens. This equipment is available since August 2008 and has been complemented by real time software to make the acquisition of articulatory data easier. We already acquired several corpus of data which are used to evaluate inversion algorithms in a first time and some coarticulations effects, for instance those corresponding to pharyngealization in Arabic. Other acquisitions are scheduled in order to collect a corpus sufficiently vast to evaluate speech recognition algorithms, and also to study speech production dynamics.
Semi-automatic speech/text alignement (ALIGNE)
An active collaboration between the INRIA Parole and Talaris teams and researchers from the ATILF laboratory has started in 2008, in the framework of the ALIGNE project of the regional CPER MISN TALC contract. The objective of this collaboration is to design and develop an interactive software to help linguistic researchers in the process of aligning speech corpora, and for the manipulation of these corpus (e.g. for the purpose of anonymisation).
The main result of this collaboration is the release of the open-source JTrans software, which is distributed in the INRIA gforge. This project, which is lead by the Parole team, should continue at least until the end of 2009, and shall probably be continued in 2010, with a shift of focus towards syntactic processing of speech.
Perception and production of prosodic contours in L1 and L2
This action, launched by the CCOSL, aims at developping collaboration between academic partners from Lorraine laboratories and universities. It has started in september 2009 and should last until the end of 2010. The speech team from LORIA is associated with the laboratory ATILF (Mathilde Dargnat). The project deals with the perception and production of prosodic contours in the first language (L1) and in a second language (L2). We have chosen two radically different languages with respect to prosody : French and English. The French intonation is characterized by recurrence of well-marked “continuation rises” whereas the English intonation is shaped by the recurrence of prominent peaks. The object of the action is to investigate the impact of a language characteristics on the perception and production of the other language. To that purpose, we will use speech processing tools developped in our team, especially those devoted to the automatic modification of prosody (fundamental frequency, durations, and energy).