Android and Speech Recognition Hacking

As part of this project, I developed an Android app that uses the built-in voice recognition to let nurses check items for a patient as he is getting ready for surgery.

Voice recognition is was found to be not good enough, for some words. We are examining possible ways of post-processing the results that come back from Google in order to improve the accuracy, for our domain. For this, I am starting with simple Bayesian learning techniques, but, we might try other techniques in the near future.

The app you see on the right was just for testing. I have another app which implements the actual checklists. We plan to tie it to the hospital's Electronic Medical Records systems using their REST API. Fun stuff!

As always, if you are a USC student (grad or undergrad) interested in hacking on think kind of stuff, just email me. We are also building 3D simulations in Unity.