Enhancing speech intelligibility based on noise characteristics

Abstract

In degraded listening conditions, speakers are known to adapt their speech via the Lombard reflex to make it more comprehensible. This characteristic has been used in previous work to modify speech recorded in quiet before it is rendered in a noisy environment. The spectral modifications used have been found to be effective in low-pass noise such as babble noise. In this work, we investigate intelligibility enhancement of speech in completely different noise characteristics, namely aircraft noise, with its dominant high-frequency components. Natural Lombard speech elicited in aircraft noise was observed to be spectrally similar to Lombard speech in babble noise and showed no intelligibility benefit in a listening test in the presence of aircraft noise. Synthetic modifications using a data dependent optimization based on a perceptual measure are investigated to obtain intelligibility enhancement in aircraft noise.