I work generally in the Speech and Audio Processing group at Idiap Research Institute, and particularly with
We work on the conjunction of speech recognition and speech synthesis. In each case, we try to focus on physiologically plausible systems. Here is a short bio.
A (possibly incomplete) list of my patents. They are mainly US, but in practice most of these began life as either GB or Japan patents; the US ones are just easier to reference.
Other technical reports that are not commercial in confidence or published elsewhere. The list includes some MPEG submissions that made it into MPEG-7.
Most work is on my GitHub page.
- For a while now I’ve been trying to find a sustainable way of maintaining software for the group. BuSLR was a failed attempt, but does contain some useful conda recipes. Right now, the most promising approach is homebrew, for which I maintain a tap.
- SSP is a signal processing library in python. It contains, amongst other things, a feature extractor and continuous pitch estimator. libssp duplicates some functionality in a C++ library.
- ISS is/was a framework for building and evaluating ASR and TTS models. Based around HTS it’s quite old.
- I wrote and (infrequently) maintain Tracter, a data-flow framework that formed the feature acquisition mechanism of Juicer. Tracter wraps several other packages, including libssp.
Software ages; Juicer and ISS were effectively replaced by Kaldi.
Since May 2019, Ina Kodrasi and I teach the course “Fundamentals of Statistics for AI” for UniDistance. The course was originally designed with David Ginsbourger. A while ago now, I held a lecturer position at EPFL, and taught a course on Speech Signal Processing.
Look at the username above and add
@idiap.ch. It’ll probably get
here. Otherwise try the
Idiap contact page.