I am currently a Master's candidate at Carnegie Mellon's Language Technology Institute.  I returned to school after being in industry for several years.  My background includes many different disciplines from particle physics and designing toys for intelligence agencies to inventing patented lighting technologies and working with statistical machine translation for low-resource languages.  I also started two companies: one for my consulting services and the other for working on interesting ideas with a few friends with the hopes of one day making a profit.  I received my bachelor's in Electrical Engineering from Virginia Tech with a minor in Mathematics.

Although I find my interests evolving everyday, my current interests at Carnegie Mellon include unsupervised and semi-supervised techniques for audio and speech analysis. I am also interested in exploring alternative approaches to learning phonemes using reinforcement learning that is inspired by early language acquisition in children.   

Previously at Carnegie Mellon, I worked with speech recognition for Arabic dialects using the RADC corpus.  My research during my first year at Carnegie Mellon involved methods for augmenting and improving acoustic models by automatically discovering rules that govern the pronunciation of common words in different dialects of Arabic.  The idea has shown potential, and can be pursued to benefit the Arabic automatic speech community. Specifically, I can see this leading to methods for incorporating the years of effort the community has invested Modern Standard Arabic into the current dialectal Arabic systems. 

I also worked on a government sponsored project to produce the next generation of analytical and information retrieval tools for unconstrained videos.  I primarily worked on classifying/detecting the topic of a video based on aural semantic features. I also worked on music analysis.

This semester, I am excited to start a new vein of research with Rita Singh and Jim Baker. I will be exploring alternative methods for speech and acoustic modeling. This includes new ways of framing the statistical modeling problem via semi-supervised learning.