Real time processing of VoIP Speech for Spoken keywords

The goal in this project is to develop a robust keyword spotting system for VoIP speech in real time which involves several unique challenges that demand very fast processing of concurrent sessions, high accuracy and minimal false alarms in the outputs, handling an unrestricted vocabulary and robust performance to codec and channel variations.

Useful Links

Link to sourceforge discussion forum

Updates


A time-lapse video of running s4 setup of TIMIT using Kaldi
An online demo of the Kaldi recognizer