2010

Research Papers

1. •S. Gulati, P. Rao, “Rhythm Pattern Representations for Tempo Detection in Music”, in International Conference on Intelligent Interactive Technologies and Multimedia, Allahabad, Dec. 2010, India    pdf
   

2. •S. Kini, S. Gulati, P. Rao, “Automatic Genre Classification of North Indian Devotional Music”, in National Conference on Communications, Bangalore, Jan. 2011, India    pdf
   

3. •S. Gulati, V. Rao, P. Rao “Meter detection from audio for Indian music”, Proc. of International Symposium on Computer Music Modeling and Retrieval (CMMR) and Frontiers of Research on Speech and Music (FRSM), Bhubaneswar, March 2011, India    pdf
   

Rhythm accompaniment

This is an application that I have started working on personally. As the name suggests, the idea is to provide a rhythm accompaniment. But, the key point is not to use metronome (as many people are not very familiar with setting BPM correctly, 60/120? kind of doubt, to make it extremely handy), which is being used by many accompaniment app. As a result, tempo estimation in correct octave and phase estimation becomes crucial. My idea is to set it up real-time, for music which requires simple repetitive rhythm accompaniment. (like bhajans in Indian Music, to start off with).  

Signature phrase extraction

In a nut shell this project aimed at finding the number of times each phrase has been repeated in a song and then picking up the most frequently occurring one as the Key phrase or Signature phrase of the song. I used chroma features to compute similarity matrix and subsequent stages of basic image processing to extract the structure of the song. Extracted phrase is to be used by Query-by-Humming system.  

Rhythm Analysis Toolbox (RATB)

I am developing this toolbox in C for rhythm analysis of music, Eg. extracting rhythmic descriptors: tempo, tatum, meter etc and many low level rhythmic features from an audio signal. I hope this will help my colleagues in their work in MIR. [Can’t share code/exe]

2009

Voice Quality Enhancement Devices

While I was in Anveshan Telecom Pvt. Ltd. I worked on Acoustic Echo Cancellation and Automatic Level Control modules for mobile network. As the deployment of these devices was intended for network side, handling heavy non linear distortion introduced by speech codecs (AMR) on the way was a challenge. This project furthered my knowledge on adaptive filtering and designing feedback/feedforward loops. All the coding was done in C.

Voice-band FAX MODEM and T.30 Protocol

This project involved designing carrier recovery module, channel equalizers and jitter control loops for voice band FAX MODEMs. Development was done in Simulink and final implementation in C followed by optimization of the code. Working independently I also implemented FAX T.30 protocol in C. (Again@ Anveshan Telecom)

2008

WMA decoder optimization

While working in ITTIAM Systems Pvt. Ltd. I performed platform specific optimization of audio codecs. I worked on WMA decoder for its optimization on a vector processor ARM cortex-A8. Good amount of knowledge about the architecture of the processor was required for this work as most of the optimization  techniques exploit pipeline structure.

Ornithopter-mechanical bird

This was a project done as a part of the elective course on Computer Aided Eng. Design offered by mechanical department. It was my sheer passion for engineering and mechanical systems that I opted this course. Though it required simulation of the mechanism for converting rotatory motion to linear, most of the tricks which worked were based on trials and heuristics. It was built using all household stuff and balsa wood.  

Audio Diarization

The goal of this project was to classify segments of broadcast news to speech, music/advertisement and silence/noise. I used HTK to build GMM models for these three categories based on MFCC features and subsequently classified the audio stream. This was my first exposure to machine learning.    Report

Guitar distortion pedal

Course project for a masters level course on architecture of digital signal processors. I implemented digital guitar tuner and distortion pedal on Texas Instruments chipset. The effects implemented were overdrive, reverb, echo, flanger, wah-wah. It was very satisfying to play guitar on self designed distortion kit.

Digital color restoration of faded motion pictures

Term paper for a masters level course on video processing. Using color correction matrices and histogram manipulation techniques, the image colors were enhanced. Below are the images that resulted from the implemented system.   Report

                                   Original image                                              Image after histogram stretching

                   

2007

Panoramic Image Mosaicing

This was a term project for EE301 Image processing course. The idea was to collect images clicked from different perspective and stitch them together to form single panoramic image. The input to system are two images with some overlap in them and four manually marked points on both images which are corresponding to each other. Based on this input an appropriate projective transform is calculated which minimizes an error function.  

Low bit rate image coder

This term paper was a part of course work on Advance topics in digital filtering. Simulation of a low bit rate image coder was done using differential predictive coding and multi rate processing in MATLAB.  Report