Prafulla Kalapatapu

• As CO-PI, submitted UGC Major Project (20 Lakhs). 
• Guiding ME student’s projects in the area of Data Mining, Information Retrieval and help them strengthen their skills and knowledge base.
• Excellent time management skills to ensure targets are met and plans completed efficiently.
• Very focused and passionate about Research
• Willing to take responsibility for the quality of research promoted and to make necessary changes to improve quality and maintain standards.
• Able to organize conferences and establish contacts with the wider academic community.

SEMINARS & WORKSHOPS (on Music Information Retrieval)

• Attended the Doctoral Consortium at International Conference on Advanced Computing Networking and Security-2011 (ADCONS 2011) held at NITK, Surathkal from Dec 16th to 18th 2011.
• Attended the 3rd CompMusic workshop held at IITM, Chennai, India from Dec 13th to 15th 2013.
• Attended a one-week Short-term Course on "Selected Topics in Data Science (STCDS- 2015)" from 14th Dec to 19 Dec, 2015 at NIT Rourkela.
• Delivered a talk on "Voronoi audio similarity for Music Information Retrieval" in Data Science meet-up on 18th Dec 2015, held at NIT, Rourkela.

Projects:

• Smart Playlist Generation using Data Mining Techniques: In this work we attempt to suggest similar songs based on seed songs selected by the user. Hybrid filtering techniques are used to find similarity between the seed songs and songs in the training set. Technologies: R, MatLab, Java. 
• Playlist generation based on user perception of songs: This work supplements such existing systems by providing user perception of songs conveyed in Twitter messages. The proposed system combines audio based features and sentiment associated with the song. This unique coupling results in play list that contains songs that are trendy with high sentiment score and to bridge the semantic gap that prevails between audio based description of song and the users’ perception. Technologies: Topsy Analysis, Java. 
• A Study On Feature Selection And Classification Techniques Of Indian Music: In this work we have reported the effect of feature selection on the accuracy of genre classification on Indian music for three genres under study. Our experimental results prove that feature selection does not always improve the classification accuracy. Hence one must employ proper evaluation methods to understand the effects of feature selection and also the selection of the right classifier. Two most significant observations of this study are that information gain based feature selection gives better and consistent accuracies than other feature selection algorithms and neural network and SVM classifiers are the best suited classifiers for Indian Song dataset. Technologies: jAudio, Matlab  MIR toolbox, Java. 
• A novel similarity measure: Voronoi Audio Similarity for Genre classification: We have proposed a new approach to measure similarity namely Voronoi Audio Similarity (VAS) by using content-based features and have investigated its effectiveness on application to genre classification. We inspected the songs at the frame level by extracting the features described in sections. The proposed approach was a combination of unsupervised and supervised method for classification. We used K-Means to cluster the frames that are similar to each other, and the process of learning is identifying the decision boundaries. These boundaries will be of use for classification purpose. The empirical results and analysis revealed that the proposed similarity measure is efficient and can be used in the domain of Music Information Retrieval(MIR) for comparing the distance between songs. Technologies: jAudio , Matlab  MIR toolbox, Java. 
• Genre similarity and song speech discrimination using a single acoustic feature: This project seeks to find if there exists a single acoustic feature(Pulseclarity) which has a very small range of values for all the songs and can distinguish between speech and music with almost 100% accuracy and there is a well-defined separation between the pulseclarity values of songs and speech. Technologies: jAudio , Matlab  MIR toolbox, Java. 
• Self Organising Maps for Music Recommender Systems: The aim of the project is to be able to identify the distinguishing characteristics of genres of Indian music and use them to select songs based on user choice by making use of Self Organising Maps(SOM). By comparing the tracks selected by the user with these features, we hope to select the nearest neighbours of these songs. In doing so, we ensure similarity to the original seed song is maintained. We also introduce serendipity by recommending songs of a different genre but similar in features. The recommender system will retrieve songs similar in features to the results of the trained dataset. We can safely conclude from our results that Bollywood and Ghazals share a common structure however Indian Classical Music is vastly different from these two genres. This generalization will help in recommending songs to the user as we have identified the general patterns that songs can have. Technologies: R, Matlab MIR toolbox, Java.

Teaching:

• Overall Experience in teaching is 14 years
• Being aware of curriculum developments and industry developments.
• Ability to maintain high standards of achievement, behavior, discipline and punctuality amongst students.
• Involved in training to students for placing into jobs.
• As an Instructor in-charge for large section courses like Object Oriented Programming, Computer Programming, Multimedia Computing with a consistent good student feedback.