Robust Speech Recognition:
Field deployment of speech recognition technology results
in a number of interesting problems, such as microphone
saturation, which severely limit the performance of speech
recognition engines. In this project, we are developing
algorithms that can postprocess the speech signal before
feature extraction, and can mitigate the effects of
saturation, clipping, and other forms of signal degradation.
At the core of this research is an interesting information
theory question of whether you can recover lost information
in a speech signal using the redundancy in the signal.
Acoustic Model Adaptation:
This project also focuses on improving the performance of a
speech recognition engine by adapting the speaker independent
HMM-based acoustic models to speaker dependent acoustic
models. Techniques to map models based on discrete distributions
to continuous distributions are being investigated.
This project is funded by
Conversational Computing Corporation (Conversay).
Conversational Computing Corporation is a Redmond, WA based
developer of speech technology solutions for both mobile and
traditional internet access devices.