Research
My research interests are speech processing, statistical estimation (eg Machine Learning). Other fields of interest are musical audio processing, and other musical related scientific problems (interfaces, computer languages for music processing and representation, etc...), although I have not done any research on those fields.
Current research areas
Voice Activity Detection
Since April 2006, I am a PhD student under the supervision of Pr. Kawahara. Right now, I am working on the problem of Voice Activity Detection, that is detecting speech boundaries in audio signals. I am particularly interested in real-time methods for speech detection in human to human interaction situation, where the speech flow changes during time. You can find some articles on the work I am doing there: my publications.
Sequential learning
While trying to solve the VAD problem, I've become more and more interested in online learning methods such as online EM and derivative, as a way to adapat sequentially to the data without a need to train the model first. I am working on applying those techniques to speech related problems. More on sequential learning.
Scientific computing
Because code produced by researchers is not just implementation but also a way to communicate ideas to other people, I have also started to look at implementation problems related to those fields (computer language for numerical computation and representation, etc...). As such, I am becoming more and more involved with scipy, a serie of tools for doing numerical computation in Python. Because of it has a really clear syntax, while being powerful enough to enable high level concepts, I believe Python to be an excellent tool for research in machine learning and signal processing. I hope that once it has matured, it will become a viable, open-source alternative to matlab.
I have implemented a fairly complete toolbox for estimating mixtures of Gaussian using Expectation Maximization (for maximum likelihood as well as Variational Bayes method for a Bayesian approach to Gaussia mixture models estimation), and am now working on LEARN, a machine learning toolkit for the scipy environment. I also developed some small tools useful for audio signal processing, such as audiolab, which can be used to read/write a large variety of audio files directly into numpy arrays, as well as pysamplerate, a simple wrapper around SRC, for high quality sampling rate convertion.
Bibliography
A beginning of a publications' list can be found there: my publications
Interesting Links
In this section, you will find links which are of some interest for my own research. I keep them here for myself, but they may be useful for others.