The aim of this project is to learn how to make predictions on what genre of movie a user is likely to be interested in based on their raw Facebook data. We collaborated with other groups to collect a dataset of 10k unique users which was then parsed, stemmed and tokenized. We trained two predictors each using a di erent model, SVM and Naive Bayes. In order to correct for the skew in the traini...

In this paper an approach minimizing the human involvement in the manual annotation of speakers is presented. At each iteration a selection strategy choses the most suitable speech track for manual annotation, which is then associated with all the tracks in the cluster that contains it. The study makes use of a system that propagates the speaker track labels. This is done using a agglomerative ...

Recent perception experiments confirm that listeners are clearly able to distinguish between professional speaking styles, even when semantic content is removed by low-pass filtering the original speech signal. The objective of this study is to evaluate whether acoustic metrics that characterize the overall fundamental frequency contour also vary significantly according to speaking style. The a...

This paper deals with a presentation of an experimental system (called WebSubDub) for creating a high-quality alternative audio track for TV broadcasting. The system is used to create subtitles for TV shows in such a format which allows to automatically generate an alternative audio track with multiple voices employing a specially adapted TTS system. This alternative audio track is intended for...

A prototype system for implementing a virtual Personal Aerobics Trainer (PAT) is presented. Unlike workout video tapes or TV exercise shows, this system allows the user to create and personalize an aerobics session to meet the user's needs and desires. Various media technology and computer vision algorithms are used to enhance the interaction of the character by enabling it to watch and talk to...

Traditionally image reconstruction is done by performing Fast Fourier Transform (FFT). But recently there has been growing interest in using compressive sensing (CS) to perform image reconstruction.In compressive sensing, the main property of signal-Sparsity is explored for reconstruction purposes.In this paper, for image reconstruction, various optimization techniques like L1 optimization, Tot...

Computer Vision and machine learning methods were previously used to reveal screen presence of genders in TV and movies [4]. In this work, using head pose, gender detection, and skin color estimation techniques, we demonstrate that the gender disparity in TV in a South Asian country such as Bangladesh exhibits unique characteristics and is sometimes counter-intuitive to popular perception. We d...

Technology used in Digital TV has the potential to enhance the viewing experience for millions of hard of hearing people. The Clean Audio project commissioned by the Independent Television Commission (ITC), and continued by Ofcom, looks at methods by which the extra information contained in 5.1 surround sound broadcasts may be used to improve the intelligibility and enjoyment of television audi...

This research blends machine learning-based discovery of preference patterns that uses natural language processing for TV viewing data with explanatory modeling that uses econometrics, as a basis for understanding TV viewing preferences at the household-level. We employ a dataset of about 1.1 million observations that was collected via set-top box technology that tracked household-level consump...

Sequences found at the beginning of TV shows help the audience absorb the essence of previous episodes, and grab their attention with upcoming plots. In this paper, we propose a novel task, text recap extraction. Compared with conventional summarization, text recap extraction captures the duality of summarization and plot contingency between adjacent episodes. We present a new dataset, TVRecap,...