You are here

Adaptation to speaker and environment

In distant-talking scenarios it is common to deal with a variety of different acoustic environments:rooms of variable size, non-stationary noises, moving speakers.

As such, a recognition system should be able to rapidly adapt to the given acoustic condition, using a limited amount of acoustic information estimated from the target environment. The adaptation procedure can exploit dynamic information of room acoustics, integrating information from other components (e.g. acoustic event localization/classification, speaker identification) and improving more general algorithms such as MAP or MLLR.