An Evolutionary Confidence Measure for Spotting Words in Speech Recognition

Confidence measures play a very important role in keyword spotting systems. Traditional confidence measures are based on the score computed when the audio is decoded. Classification-based techniques by means of Multi-layer Perceptrons (MLPs) and Support Vector Machines have shown to be powerful ways to improve the final performance in terms of hits and false alarms. In this work we evaluate a keyword spotting system performance by incorporating an evolutionary algorithm as confidence measure and compare its performance with traditional classification techniques based on MLP. We show that this evolutionary algorithm gets better performance than the MLP when False Alarm (FA) is high and always performs better than the confidence measure based on the single score computed during the audio decoding.