Detail

Multimodal speech algorithms and applications

In this talk I cover three of the latest research projects I have been
leading in Telefonica over the last months. The first project, titled
“spoken wordclouds”, uses pattern-matching algorithms to automatically
discover acoustic repetitions in speech recordings and then cluster them
to obtain a summary of the recording, in a similar way to what a wordcloud
does with text. In the second project I present the efforts I am leading
in the field of multimodal video-copy detection, used for example, for
detecting the infringing usage of copyrighted multimedia material. Last,
the project called “spoken ebooks” proposes a method to synchronize an
ebook with its corresponding audiobook and then be played in synchrony to
the user. Time permitting, I will be showing a live demo of this project
in an Ipad.