Abstract: We describe a computer application for relaxation that is based on autonomous music generation synchronous to user's actions. We provide the results of a user study, showing that though the application is relatively simple, it is perceived as computationally creative by its users. The paper summarizes our observations on both computers' and human creativity that were made in process of implementing the application.
Poetry Framework Tool
Abstract: In this paper, we present the exploratory work we did in identifying the creative aspects of creating and understanding poetry. Systems exist that are able to generate poetry given several key input parameters: such as the theme, a source corpus of text, rhyme scheme, tone of the poem (comedic, serious etc.,) and so on. We have tried in implementing a system which just requires a source corpus and a rhyme scheme. We have found that although the poems generated by our technique are not very sophisticated, they reveal important differences between human and computational creativity; especially the limits of a computer's ability to “create''. We have also created a system which could identify poems within text, as well as a creativity assistant tool, which could help a novice poet in the poetry writing process.
Mood Player C#
Abstract: Music has useful social and psychological functions for individuals. In order to make use of this power, we designed a music player, i.e., Mood Player Alpha. It aims to predict moods of songs by their lyrics based on latent Dirichlet allocation. It models moods as distributions over key words in the lyrics. As long as we have the distribution and key words extracted from a lyric, we can predict the mood associated with the lyric. In order to test the idea, 200 lyrics were collected with regard to four different moods, i.e., happy, sad, angry, and relaxed. Results showed that the prediction accuracy is 66% for 10-fold cross-validation. Furthermore, a user interface was designed and evaluation in terms of creativity from 18 users was also reported.
CAPTIS: Intelligent Agent for Interactive Exercise
Abstract: Rapid progress in technology, especially in the area of articial intelligence, offers tremendous possibilities for innovation in therapy for individuals with autism or physical impairments. Clinical studies of robot-assisted therapy have demonstrated the viability and usefulness of robots in motivating patients in physical therapy and in interacting with individuals with autism. In this paper, we present the design of a interactive articially-intelligent system named CAPTIS, which uses a Kinect interface to perform similar actions to the robots used in clinical research but at a much reduced cost. The system guides a user through exercises by evaluating the user's performance and adapting to the user's preferences and performance. CAPTIS's ability to perform the three basic functions of therapy - eliciting behaviors; modeling, teaching, and/or practicing a skill; and providing feedback - at a fraction of the cost of robots reveals that CAPTIS is the beginning of a new direction in interactive AI systems for therapy
Absract: In this paper we describe the progress of the House Power Project, a solution to saving power.
Music Player Alpha
Abstract: In this paper, we propose, design and implement an autonomous, music mood-classification application, drawing on contemporary research in the field. A key aspect of our project is the leveraging of physiological studies that have been carried out for generating representative mood models. Such studies serve as the groundtruth for the identification of a set of “mood-tags” on which music is classified. We employ a Support Vector Machine (SVM) classifier that uses three features extracted from audio signal analysis – intensity, rhythm and timbre to classify music by its representative mood, which is subsequently mapped onto a 2-D mood plane. A novel contribution is the use of an intuitive interaction mechanic for playlist generation based on a user specified “mood-curve”.
*Projects described on this page are the intellectual property of the students who worked on them and Georgia Tech. They do not fall under the Creative Commons License.*