Enhancement of ambient speech for robot audition – Noise reduction and dereverberation

Whereas the impact of noise on speech quality and intelligibility has been studied and modeled extensively, only in the last 10 years or so have the effects of reverberation been taken into account with anything even approaching the same level of effort. However, in many real-world speech processing applications, noisy speech is only observed after being also subjected to significant reverberation. Commonly quoted examples of such applications include robot audition, hearing aids and hands-free terminals.

This talk will consider the effects due to combinations of noise with reverberation on speech quality, speech intelligibility and speech recognition performance. Using robot audition as an example, the problems addressed will first be outlined. Then, recent performance results will be given showing the state-of-the-art performance of various combinations of noise reduction and dereverberation processing. Conclusions will be drawn so as to identify research threads with strong potential with the aim of understanding what expectations are realistic of signal processing approaches to the enhancement of ambient speech.