Improvement In Automatic Speech Recognition Performance In Noisy Environments Using Time-Domain Blind Source Separation

Blind source separation (BSS) methods are generally used to separate speeches of people who are simultaneously speaking in the same room by using more than one microphone to record the speeches. However, in this study, the aim is to separate noise from the speech and therefore improve speech recognition performance using a time-domain BSS method. This method uses time-domain second-order statistics and is based on non-stationarity and non-whiteness properties of speech signals. Sphinx, which is a speaker independent, continuous speech recognition engine, is used to test the recognition performance of resulting enhanced speech. The simulation results demonstrate that, there is a remarkable improvement in recognition performance in terms of sentence error rate and word error rate.