Institution: Imperial College
An audio recording contains the sound of the space as well as the intended source. The aim of this project is to develop methods which extract characteristic features of the recording environment and to evaluate how effective these features are in identifying where the recording was made.
Recordings which are subjected to forensic analysis are generally expected to be monaural. In Moore2013 we consider the possibility of estimating the dimensions of a room based on only a single channel impulse response.
In Moore2013a we introduce the concept of a roomprint and discuss the types of information which might be useful. We further propose 1/3-octave band T60 across the speech band as a potential feature and demonstrate that it is very effective at discriminating between rooms.