Ressources.ircam

To gain in time and quality in movie post-production process there is a need of automatic or semi-automatic (i.e., user assisted) tools that allow modifying both audio and video material according to particular needs. In this talk we will present recent research in Technicolor for developing such tools. First, concerning video modification/editing tools, we will briefly describe research works on video manipulation conducted in collaboration with academic partners, namely object removal and face retargeting. Second, as for audio modification tools, we will mostly speak about source separation that allows extracting particular sound sources from audio tracks, thus making possible their further modification, replacement or remixing. More specifically, this may be useful for redubbing, audio remastering or audio upmixing. We will concentrate on so-called informed source separation methods that use some additional information to help the separation process.
Indeed, these methods usually lead to a superior separation quality as compared to the blind (or non-supervised) methods.
We will speak about user-guided approaches based on some annotations provided by user, about newly introduced on-the-fly audio source separation concept and about text-informed source separation. The presentation will be supplied with examples of results and demo videos.