Language identification in noisy environments using throat microphone signals

Abstract

Automatic identification of a language in a noisy environment is a challenging task. Performance of a language identification system depends on the quality of the input speech signal. In the presence of high levels of background noise, speech signals recorded using a close-speaking microphone are degraded. In contrast, a throat microphone picks up high quality speech unaffected by the surrounding noise. This paper explores the possibility of using throat microphone speech signals for text-independent language identification in noisy conditions. Languages are modelled using autoassociative neural networks based on the vocal tract system features and excitation source features derived from the throat speech signal. The results of this study show that the throat microphone speech-based language identification system performs well in noisy environments.